Chabad Torah Leining — Research Report for a Lein-Practice Tool
Date: 2026-11-17 · Scope: Torah only, Chabad (Lubavitch) trop, practice tool (pick a parsha/aliyah, hear it leined correctly in Chabad trop, follow along, repeat) · Research only — no building.
How this was made: four parallel research passes (taamim system, Chabad melody, text data, existing tools + pedagogy) plus direct verification of the voice-technology fork. Every claim below is sourced; where documentation is thin or only exists as recordings, that is stated plainly. The Chabad melody deep-dive (full write-up, motif transcription, sources) lives at ~/workspace/research/chabad-trop-research.md.
0. Bottom line
- The grammar is fully documented; the melody is not. The taamim system — which mark, what it means syntactically, how phrases parse — is written down precisely enough to implement from text. The Chabad melody itself exists only in recordings (R' Michoel Slavin, baal korei of 770, on Chabad.org) and inside TropeTrainer's "Chabad" voicing system (sheet music behind a free demo / subscription). There is no published, note-for-note "Chabad trop" chart.
- The best working model for the melody is an Eastern Ashkenazi sheet-music chart (Schmidt, Agudath Achim, Altoona PA — Conservative, not Chabad-labelled), transcribed note-by-note, matching the academic analysis (Malin 2016, after Kalib 2002 and Portnoy & Wolff 2000) at every structural checkpoint. It is a solid hypothesis for the Chabad melody — Chabad is in the same Eastern Ashkenazi (Polish-Lithuanian) family — but it must be validated against Slavin's recordings before being called Chabad.
- The text data is solved. Sefaria's API serves the whole Chumash with taamim (public domain), Hebcal has aliyah divisions as open data, and the Unicode encoding rules are known — including the traps (never NFC-normalize; glyph collisions; the zarqa/tsinnorit swap).
- The voice is the open problem, and it splits two ways. No TTS in the world sings taamim today. But a tiny-voca-style model can be driven to sing: a Nov 16, 2026 write-up verifies tiny-voca accepts per-phoneme duration and per-frame pitch overrides and sings within ~2 cents of the notes. And small Hebrew TTS exists (Phonikud/Piper, 32M params). So a synthetic Chabad voice is plausible — it needs a validated Chabad motif table and an Ashkenazi-pronunciation Hebrew voice, neither of which exists in open form today. A recorded baal koreh is proven, authentic from day one, and the whole Torah is finite (54 parshiyot, recordable once) — but the word-level alignment layer doesn't exist as open data and would have to be built.
- The gap in the market is a Chabad-first teacher, not a player. Chabad.org's trainer is free and authentic but only plays recordings. TropeTrainer is the best generalist (Chabad is 1 of 79 systems). Trope on the Go has the best pedagogy (pitch feedback, spaced repetition) but is Ashkenazi-only. Nobody teaches the Chabad system clause-by-clause with phrase looping, lookalike drills, record-and-compare with pitch feedback in the Chabad melody, and a tikkun-to-STaM progression.
1. The taamim system from first principles
1.1 What the marks are — three jobs at once
Hebrew ta'amei ha-mikra (טַעֲמֵי־הַמִּקְרָא), singular ta'am = "taste / sense / reason", Yiddish trop, English "accents / trope symbols". Fixed in Tiberias by the Masoretes, 8th–10th c. CE, culminating in Aaron ben Moses ben Asher; the Aleppo Codex (c. 920) is the model text. The marks are not in the Torah scroll — the reader learns them from a printed text (tikkun/Chumash) and chants from memory. Unicode: U+0591–U+05AF, plus sof pasuq U+05C3 and paseq U+05C0.
Every mark does three things at once:
| Function | What it does |
|---|---|
| Syntax / punctuation | Divides a verse into clauses and phrases, like commas, semicolons, full stops. Can affect exegesis — e.g. Isa. 40:3, where zakef katan on qore discourages the KJV grouping "voice crying in the wilderness". |
| Phonetics / stress | Most marks sit on the consonant of the stressed syllable, showing milra (final-syllable stress) vs mil'el (penultimate). This also shows where the main note of the motif goes. A few always go on the first/last letter, for musical reasons or to disambiguate identical shapes. |
| Music | Each mark stands for a short melodic motif (niggun / neginot in Hebrew, trop in Yiddish). Strung together they make the verse's chant. |
Mechon Mamre warns the syntax tree is not a pure syntax tree: virtually every verse is split by etnachta even when that falls mid-sentence (Gen. 1:1, between verb/subject and object). Do not assume the parse is a pure grammar tree.
Scope note: this is the 21-book prose system. The 3 poetic books — Psalms, Proverbs, Job (Sifrei Emet, אמ״ת) — use a different system with different functions for many of the same-looking signs. For a Torah-only tool this doesn't matter, but don't reuse the Torah melody table for Tehillim.
1.2 Disjunctives vs conjunctives
Rule: in general each word has one mark. Exceptions: a maqqef-hyphenated group counts as one word (one mark for the group); a long word may carry two (a disjunctive on the stressed syllable plus the related conjunctive two syllables before, in place of meteg).
- Disjunctive (mafsiq, "divider"; traditionally "kings/nobles") — marks a pause/division after that word. Disjunctives divide a verse into phrases; within a phrase all words except the last carry conjunctives.
- Conjunctive (mesharet / meḥabber, "servant") — joins the word to the following word, like a slur in music. Lowest dividing power: it terminates a unit of one word only.
Disjunctives by level
| Level | Hebrew | English | Function |
|---|---|---|---|
| 1 — Emperors | סוֹף פָּסוּק sof pasuq; the accent on the last word is סִלּוּק silluq | "end of verse" / "taking leave" | Last word of every verse. Silluq is the accent; sof pasuq (׃) is the punctuation mark after it. Do not conflate them. |
| 1 | אֶתְנַחְתָּא etnachta / atnach | "rest / pause" | Divides verse into two halves (cola). At most once per verse; a very short verse may have none. |
| 2 — Kings | זָקֵף קָטָן zakef katan | "small upright" | Usual 2nd-level disjunctive; roughly comma/semicolon. |
| 2 | זָקֵף גָּדוֹל zakef gadol | "great upright" | Replaces zakef katan when it constitutes a phrase on its own; longer tune, stronger separation. |
| 2 | טִפְחָא tifcha (Seph. tarcha) | "hand-breadth" | Replaces zakef katan in the immediate neighbourhood of sof pasuk or etnachta — the "near companion" of the emperors. |
| 2 | סֶגוֹל segol / segolta | "bunch of grapes" | Stronger 2nd-level disjunctive, used in very long verses; only in first half of verse. |
| 2 | שַׁלְשֶׁלֶת shalshelet | "chain" | Replaces segol when it constitutes a phrase on its own. Only 4× in the Torah: Gen. 19:16, 24:12, 39:8; Lev. 8:23, always at verse beginning. |
| 3 — Dukes | רְבִיעַ revia / revi'i | "fourth" | Usual 3rd-level disjunctive. |
| 3 | זַרְקָא zarka | "scatterer" | Replaces revia in vicinity of segol; only ever before segol. |
| 3 | פַּשְׁטָא pashta | "stretching out" | Replaces revia in vicinity of zakef; precedes zakef katan. Same glyph as kadma (U+0599) — distinguished by placement: pashta on last letter (doubled if word is mil'el), kadma on stressed syllable. 5,429× in Torah. |
| 3 | יְתִיב yetiv | "sitting" | Replaces pashta/kadma when it constitutes a phrase on its own; always on first letter of word. |
| 3 | תְּבִיר tevir | "broken / downward tumble" | Replaces revia in vicinity of tifcha; precedes tifcha. Frequently paired with darga. |
| 4 — Counts | פָּזֵר pazer | "lavish / strewn" | Cluster near beginning of a long half-verse; elaborate realisation. |
| 4 | גֵּרֵשׁ geresh / azla-geresh | "expulsion" | See kadma v'azla below. Doubled unless on a non-finally-stressed word or following kadma. |
| 4 | גֵּרְשַׁיִם gershayim | "double geresh" | Replaces geresh when not preceded by azla AND word is stressed on last syllable. |
| 4 | תְּלִישָׁא גְדוֹלָה telisha gedola | "great detached" | Detached — never linked to following note as one musical phrase. |
| 4 | מֻנַּח לְגַרְמֵהּ munach legarmeh (+ paseq ׀) | "munach by itself" | Munach + paseq becomes a disjunctive, mainly before revia, occasionally before pazer. |
| 4 | קַרְנֵי פָרָה qarne farah | "horns of a cow" | Once in the Torah: Num. 35:5, preceded by yerach ben yomo. |
Conjunctives ("servants", meshartim)
The general conjunctive is munach; which one appears depends on the following disjunctive:
| Hebrew | English | Precedes (usual) |
|---|---|---|
| מֻנַּח munach | "resting" | General/default; before etnachta, zakef, segol, revia etc. Can repeat many times in a row. |
| מֵרְכָא mercha / ma'arikh | "lengthener" | Before sof pasuk and tifcha (occasionally tevir and other disjunctives). Note: misleadingly translated "comma" in modern usage — it is conjunctive; a comma marks a break. |
| מַהְפָּךְ mahpach | "turning round" | Before pashta. Cannot appear without a pashta (but pashta can appear without it). |
| דַּרְגָּא darga | "stairstep" | Before tevir. Before revia, the penultimate munach is always a darga. |
| קַדְמָא kadma | "precede" | Before geresh and some conjunctives; first member of kadma v'azla. Same glyph as pashta. |
| תְּלִישָׁא קְטַנָּה telisha ketana | "small detached" | Before some conjunctives; 3rd conjunctive before pashta in the pashta chain. |
| מֵרְכָא כְּפוּלָה mercha kefula | "doubled mercha" | Occasionally preceded by darga; function similar to tevir. Only 5× in Torah: Gen. 27:25, Ex. 5:15, Lev. 10:1, Num. 14:3, Num. 32:42. Disputed whether conjunctive or occasional replacement for tevir. |
| יֶרַח בֶּן יוֹמוֹ yerach ben yomo | "day-old moon" | Before qarne farah (once in Torah). |
Count disagreement — flag: sources count differently. Iancu (2025) says the Ashkenazi system has 27 marks: 19 disjunctives + 8 servants. Wikipedia's table lists ~27 prose rows plus poetic-only rows. Any implementation should count codepoints, not "names", and treat pashta/kadma, geresh/azla, mahpach/yetiv as glyph collisions resolved by position/context, not by codepoint alone.
1.3 Hierarchy, domains, parsing a pasuk
The ruler model (Mechon Mamre — most formal source). Classes carry ruler titles: qeysar (emperor) terminates a whole verse and "reigns" it; a melekh (king) divides an emperor's realm; a mishne (duke) divides a king's realm; a shalish (officer) divides a duke's realm; lowest is mesharet (servant), a conjunctive. Exception: in the 21 books the first division is by an emperor, etnachta, dividing another emperor's realm (sof pasuk) — treat each half-verse (ending in sof pasuk or etnachta) as the top level, remembering sof pasuk divides stronger than etnachta.
Continuous dichotomy — how the tree is built. Start with a half-verse ending in an emperor → split by a king → each half split by a duke → each quarter split by an officer, etc. Key rules:
- Of two marks of equal rank with no stronger mark between, the earlier is stronger (it dominates the later ones).
- A distinctive mark M2 following M1 may be arbitrarily lower in rank than M1, but at most one rank higher — and that case occurs iff M1 belongs to M2's realm.
- Ranks, not nesting levels, are the better input for music, because ranks change smoothly while nesting jumps.
The companion table (Wikipedia "Sequences") — the last subdivision in a phrase must be a disjunctive one level down, chosen to fit the governing disjunctive (its "near companion"); other subdivisions are the "remote companion". A disjunctive constituting a phrase on its own may be substituted by a stronger disjunctive of the same level (the "equivalent isolated disjunctive").
| Main disjunctive | Preceding conjunctive(s) | Near companion | Remote companion | Equivalent isolated |
|---|---|---|---|---|
| Sof pasuk | Mercha | Tifcha | Zaqef katan | — |
| Etnachta | Munach | Tifcha | Zaqef katan | — |
| Segol | Munach | Zarka | Revia | Shalshelet |
| Zaqef katan | Munach | Pashta | Revia | Zaqef gadol |
| Tifcha | Mercha; Darga; Mercha-kefulah | Tevir | Revia | — |
| Revia | Munach; Darga Munach | Munach legarmeh | Geresh, Telisha gedola, Pazer | — |
| Pashta | Mahpach; Qadma Mahpach | Geresh/Azla/Gershayim | Telisha gedola, Pazer | Yetiv |
| Tevir | Mercha or Darga | Geresh/Azla/Gershayim | Telisha gedola, Pazer | — |
| Geresh/Azla | Qadma; Telisha qetannah | — | Gershayim | — |
| Telisha gedola | Munach | — | — | — |
| Pazer | Munach | — | — | — |
| Qarne farah | Yerach ben yomo | — | — | — |
1.4 How a pasuk's melody is built
Not fixed pitch notation. The marks are ideographic / neume-like: each stands for a short melodic motif / group of notes, not individual notes. Haïk-Vantoura's reconstruction treating signs as individual scale degrees is explicitly rejected by musicologists as at odds with all existing traditions.
- One motif per symbol — per tradition/mode. "Each community re-interpreted its reading tradition so as to allocate one short musical motif to each symbol." The written signs are identical everywhere; the melodies differ by community (Ashkenazi, Sephardi/Mizrahi, Yemenite, Italian, etc.) and by book and occasion within one community.
- Ashkenazi: six modes (Malin): (1) Torah normal Shabbat/weekday/festival; (2) Torah High Holidays; (3) Haftarah; (4) Lamentations; (5) Esther; (6) festive Megillot. Plus special variants/codas for Aseret haDibrot, Az Yashir, Masa'ot, a coda at the end of each reading, and a special coda at the end of each of the five books ("Hazak Hazak…").
- Motifs combine as phrases, not as isolated notes. A phrase = [conjunctive lead-in(s)] + [disjunctive motif]. The conjunctives are often flourishes leading into the following disjunctive rather than motifs in their own right. The Tiberian system invented the range of conjunctives precisely to "introduce and elaborate the main motif in longer phrases": e.g. tevir is preceded by mercha, a short flourish, in shorter phrases but by darga, a more elaborate run of notes, in longer phrases.
- What varies with context (verified): (1) tradition/mode — a renderer needs a per-mode motif table, not one global table; (2) phrase length / word & syllable count; (3) which higher mark follows; (4) final vs non-final in a chain of equals; (5) isolated vs accompanied (isolated disjunctives take the stronger substitute — zakef gadol, shalshelet, yetiv — with a longer/more elaborate tune); (6) stress position — milra vs mil'el changes placement/doubling and therefore where the motif's main note lands; (7) coda overrides at the end of a reading / book.
- Counter-examples to "one mark = one fixed tune": Yemenite cantillation has only 8 distinctive motifs and renders conjunctives in monotone; Sephardi Haftarah modes give different disjunctives the same/closely similar motifs. Motif identity is not 1:1 with mark identity in all traditions.
Implementation consequence: model as syntax_tree (from marks) → phrase segmentation → motif lookup (mark, mode, phrase-length, final/non-final, isolated?) → note sequence, with the motif table supplied per tradition/mode from audio/transcription. The text alone does not encode pitches.
1.5 Formal rules that matter for implementation
- One mark per word (§1.2); maqqef chains share one; long words can carry two.
- Near/remote companion + isolated-substitute rules — §1.3 table. A validator can check: a disjunctive may only be preceded (ignoring conjunctives) by a same/higher-level disjunctive or its near companion.
- Rank-step rule (Mechon Mamre): successive distinctive marks may drop arbitrarily but rise at most one rank. Cheap sequence validator.
- Once-per-verse: sof pasuk (by definition), etnachta (≤1), segol (≤1 as first of its rank), shalshelet (≤1), qarne farah + yerach ben yomo (1× in Torah).
- Common phrase groups (use as test vectors): (Mercha) Tifcha (Mercha) Sof-Pasuk — end of every verse; (Mercha) Tifcha (Munach) Etnachta — ≤1×/verse; (Mahpach) Pashta (Munach) Zaqef-qaton — most common; Zakef-gadol standalone; [Munach] Zarka [Munach] Segol — zarka only before segol; Munach (Paseq) [Munach] Revia; Darga Tevir; Mercha-kefula usually alone.
- Yetiv / Pashta pair: pashta precedes zakef katan; replaced by revia when a stronger stop is needed or when >3 in a row — but the last stop before the zakef can never turn into revia. Mahpach must be followed by pashta, but yetiv may be followed by pashta or even by mahpach pashta. Mahpach vs yetiv are distinguished by placement/context, and are confusable when the first accented syllable has cholam/shuruk not beneath the word.
- Kadma v'Azla: the pair kadma (conjunctive) + azla/geresh (4th-level disjunctive) — "go forward and depart". May precede a mahpach, a revia group, or a tevir group. Yacavone: treat as one accent. Wikipedia: technically qadma/azla before gerish is a different sign from qadma before other disjunctives, though they look identical; Sephardim reserve azla for the former. Do not merge them in a parser without a flag.
- Pazer / Telisha cluster: 4th-level marks cluster, usually in order pazer → telisha gedola → geresh/gershayim, near the beginning of a long half-verse.
- Placement / glyph collisions (critical for tokenisation): - Pashta vs kadma: same glyph U+0599; pashta on last letter (doubled on mil'el words), kadma on stressed syllable. - Geresh vs azla: same glyph U+059C; geresh doubled unless mil'el or after kadma. - Mahpach vs yetiv: similar shape; mahpach under first letter of accented syllable, yetiv under first letter of word (always first letter). - Munach vs silluq/meteg: silluq looks like meteg (U+05BD); meteg is not a cantillation mark. - Zarka vs tsinnorit Unicode swap: encode zarqa/tsinnor as U+05AE "HEBREW ACCENT ZINOR", and tsinnorit as U+0598 "HEBREW ACCENT ZARQA" (Unicode Technical Note #27). A naïve name→codepoint map will be wrong. - Paseq (׀, U+05C0) can make an otherwise conjunctive mark distinctive (munach legarmeh), or mark an independent division. - Placement: pashta, zarka, segol, telisha always on leftmost letter; yetiv and telisha gedola always on rightmost letter (modern texts often duplicate/add a related accent on the stressed syllable to aid pronunciation).
- Exceptions are numerous: Mechon Mamre: "for each of these rules there are numerous places in the Hebrew Bible where they are violated." A validator should warn, not reject, on violations.
- Corpus sanity checks: Etnachta 5,483× in Torah; Pashta 5,429× in Torah (Price concordance, via Wikipedia).
Implementation sketch (derived, not a source claim): 1. Tokenise by Unicode codepoint (U+0591–U+05AF, U+05BD, U+05C0, U+05C3), resolving glyph collisions by position in word + following mark (pashta/kadma, geresh/azla, mahpach/yetiv), with the zarqa/tsinnorit swap hard-coded. 2. Parse the disjunctive sequence per half-verse into a tree using ranks (emperor 0, king 1, duke 2, officer 3, servant 4) + the rank-step and near/remote-companion rules; validate once-per-verse marks. 3. Segment into phrases = [conjunctives…] + disjunctive; each phrase is one musical unit. 4. Render via a per-mode motif table keyed by (mark, mode, phrase-length class, final/non-final, isolated?) — not by a single global pitch map.
2. The Chabad trop melody specifically
2.1 Bottom line
- No published, note-for-note "Chabad trop" chart exists anywhere found. The Chabad melody lives in recordings, not in staff notation.
- The one place it exists as a named system with sheet music is TropeTrainer (tropetrainer.com) — "Chabad" is one of its 79+ voicing systems (Torah, Haftarah, High Holidays, Esther), with sheet music for every melody and a free demo, no account. That is the first source to check for machine-usable Chabad data.
- The best freely available notated chart is a 5-page Eastern Ashkenazi sheet-music PDF (Stephen Schmidt, Agudath Achim Synagogue, Altoona PA — Conservative, not Chabad-labelled). Every note of it was transcribed programmatically from the PDF plus visual verification. It matches the academic analysis of Eastern Ashkenazi Torah trope (Malin 2016, after Kalib 2002 and Portnoy & Wolff 2000) at every structural checkpoint, so it is a solid working model — but it must be validated against Chabad recordings before being called canonical Chabad.
- Ground-truth Chabad recordings: R' Michoel Slavin (baal korei of 770, the Rebbe's minyan) on Chabad.org — all 54 portions + haftarot, in the interactive Torah Trainer; and R' Mordechai Schusterman (read in the Rebbe's presence at 770) via Beverly Hills Chabad.
2.2 How the system works
- One ta'am per word at its stressed syllable; each mark = a short melodic phrase and a syntactic function. Marks are universal; melodies vary by community.
- Motifs are phrase-level, not per-mark. Tipcha is sung differently before etnachta vs. sof pasuk vs. end-of-reading; merkha/munach change shape with the disjunctive that follows. Model phrases, not single marks.
- Torah mode (ordinary Shabbat/weekday/festival) is one of six Ashkenazi modes; it is in a major scale, final = tonic, recurrent resting point on scale-degree 2, no fixed meter/beat. Chabad.org's "Who Made Up the Way We Sing the Torah?" (R' Baruch Davidson) gives the history/theology: marks from the 9th–10th c. Masoretes; the tune itself held to go back to Moses.
- There is no fixed meter or beat; trop is chanted freely. Rhythm is not part of the notation (the Schmidt chart says so explicitly: "Trope are chanted freely, not according to the note rhythms shown here").
- The Rebbe's sichos treat the ta'amim as carrying meaning beyond conscious perception; no sicha giving note-by-note melodies was found.
2.3 Chabad vs. other traditions — how big are the differences?
- vs. Lithuanian/Yeshivish (Polish-Lithuanian): SMALL. Same family — Chabad arose in Liozna/Liadi, White Russia, inside the Lithuanian area. No source found enumerates note-level differences. The distinction is recognised (TropeTrainer lists "Chabad" separately from "Common Melody", "Spiro", "Wolff/Portnoy" — 32 Torah systems), but it is motif-level variation; Malin: "There is no single authoritative source for the melodies… even within the Eastern Ashkenazi tradition." Two concrete Chabad details found: the Chabad shalshelet is "basically 2 pazers", and Slavin's etnachta is nearly monotone where older sheet music gives it two notes (Chabad.org forum — i.e. variation exists even within Chabad).
- vs. other Ashkenazi (Avery/Binder Reform, Western/German): MODERATE. Same grammar, audibly different tunes; e.g. sof pasuk ends 3–1 (Kalib) vs. 1–5 (Portnoy & Wolff).
- vs. Sephardi: LARGE. Different system: Jerusalem-Sephardic/Syrian Torah is (almost) always in Maqam Sigah; conjunctives are flourishes into the following disjunctive, not motifs in their own right; different disjunctives often share motifs; Yemenite has only 8 motifs total. Names differ too (Sephardi qadma = Ashkenazi pashta).
2.4 The Torah mode in numbers (Eastern Ashkenazi, cross-checked)
Tonic = C in the Schmidt chart (C major, no sharps/flats except one marked B♭ in the segol melisma). Scale degrees below are relative to C (1=C5, 2=D5, 3=E5, 4=F5, 5=G5, 6=A5, 7=B5; below the tonic 7=B4, 6=A4, 5=G4, 4=F4). Range of the whole chart: F4–A5. Reciting tones are 1 (C5) and 2 (D5); the final is 1. (Malin's Kalib/Portnoy examples are notated in F major with one flat — same relative degrees, different absolute pitch. Absolute pitch is a free choice; readers chant at any comfortable level.)
2.5 Motif inventory
Transcribed from Stephen Schmidt, "Torah Trope" (5 pp., treble clef, C major), https://www.agudatachim.com/wp-content/uploads/2026/05/torahsheet.pdf — every notehead measured programmatically from the PDF (staff-line geometry → diatonic pitch) and verified visually staff by staff, including beams, rests, the B♭/♮ signs in the segol staff, and the open-headed half-note final of sof k'ria. Cross-checks against Malin 2016 noted per row.
Confidence key: A = notated in the chart and corroborated by Malin/Kalib/Portnoy · B = notated in the chart only · C = known only from recordings/forum, not notated anywhere found.
| Ta'am | Motif as notated (C major; phrase context in brackets) | Degrees | Conf. |
|---|---|---|---|
| Sof pasuk (siluk) | [merkha–tipcha–merkha–] D5 D5 E5 C5 | 2 2 3 1 | A |
| Etnachta | [merkha–tipcha–munach–] C5 C5 D5 | 1 1 2 | A (but see Slavin variation below) |
| Zakef katan | [mahpach–pashta–munach–] G5 G5 E5 D5 | 5 5 3 2 | A |
| Zakef gadol | C5 C5 E5 G5 E5 D5 | 1 1 3 5 3 2 | B |
| Tipcha | D5 E5 (before etnachta/sof pasuk); D5 G5 E5 at end of reading | 2–3; 2–5–3 | A |
| Revia | [munach–munach–] E5 E5 D5 C5 B4 A4 | 3 3 2 1 7 6 | A |
| Tevir | D5 C5 B4 C5 D5; preceded by darga G5 G5 F5 E5 D5 (long) or merkha D5 D5 (short) | 2 1 7 1 2 | A |
| Pashta | [mahpach–] C5 G5 | 1–5 | A |
| Yetiv | G5 G5, then munach E5 E5 D5 + katon G5 G5 E5 D5 | 5 5 | B |
| Segol | [munach C5 C5, zarka C5 C5, melisma B♭4 A4 G4 A4 G4 F4, munach F4 C5,] C5 D5 C5 | 1 2 1 (melisma ♭7…4) | B |
| Zarka | C5 C5 reciting + descending melisma B♭4–A4–G4–A4–G4–F4 (only chromatic note in the chart) | 1 1, then ♭7…4 | B |
| Munach | E5 C5 (before etnachta); E5 E5 D5 (before katon/yetiv); C5 C5 (before zarka/telisha/segol) | varies | A |
| Kadma | C5 C5 (reciting on 1), shown only inside kadma-v'azla | 1 1 | B |
| Mahpach | G5 G5 C5 (before pashta) | 5 5 1 | A |
| Merkha | B4 D5 B4 (before tipcha); D5 D5 (before tevir / sof k'ria); E5 C5 (before sof pasuk) | 7–2–7; 2–2; 3–1 | A |
| Darga | G5 G5 F5 E5 D5 | 5 5 4 3 2 | B |
| Telisha gedola | Recite C5 ×8, close E5 D5 E5 | 1… 3 2 3 | B |
| Telisha ketana | Recite C5 ×8, close E5 D5 C5 (differs from gedola only in last note) | 1… 3 2 1 | B |
| Geresh | G5 G5 D5 | 5 5 2 | B |
| Gershayim | C5 D5 E5 D5 C5 D5 E5 | 1 2 3 2 1 2 3 | B |
| Kadma-v'azla | Kadma C5 C5; azla C5 E5 G5 E5 A5 G5 | 1 1 / 1 3 5 3 6 5 | B |
| Pazer | C5 C5 D5 D5 E5 A5 G5 E5 D5 | 1 1 2 2 3 6 5 3 2 | B |
| Sof k'ria (end-of-reading coda) | Merkha D5 D5, tipcha D5 G5 E5, merkha E5 C5, D5 E5 D5 A4 (final ah = half note) | 2 2 / 2 5 3 / 3 1 / 2 3 2 6 | A |
| Shalshelet | Not in chart; forum: "The Chabad Shalshelet is basically 2 Pazers." 4× in Torah. | — | C |
| Karnei parah | Not in chart; forum: "a combination of the Mercha Kefulah and Telisha Gedola." 1× in Torah. | — | C |
| Mercha kefula | Not in chart; 1× in Torah. | — | C |
| Yerach ben yomo | Not in chart; 1× in Torah. | — | C |
| Munach legarmeh | Not in chart as such (munach + paseq); use contextual munach figure. | — | C |
Notes on the table: - The chart notates phrases, not isolated marks — that is the correct unit. A ta'am's realisation depends on its neighbours and on position (mid-verse vs. end of verse vs. end of reading). - The chart lists phrase variants (e.g. etnachta 1A–1D, katon 3A–3F, telisha 9A–9D) but notates only the fullest form of each. - Two published variants differ from the chart at one point: Portnoy & Wolff's sof pasuk ends 1–5 where Kalib (and the chart) end 3–1 (Malin, Ex. 10a vs 10b). Treat the sof-pasuk ending as a known variation point. - The chart's home is Agudath Achim Synagogue, Altoona PA (Conservative; Dr. Stephen Schmidt, a member). It labels no tradition. It is used here as the Eastern Ashkenazi working model, not as a Chabad document. - Etnachta variation: Slavin's near-monotone rendering (Chabad.org forum, described as his personal style) differs from the chart's C5–C5–D5 — flag as a validation point, not a contradiction to resolve here. - Pronunciation: Chabad uses Ashkenazi pronunciation (e.g. tav, sav). A TTS/voice must use Ashkenazi, not Israeli/Sephardi, vowels.
2.6 What is text-reproducible vs. recordings-only
| Layer | Status |
|---|---|
| Grammar (phrase hierarchy, conjunctive/disjunctive rules, recursive dichotomy) | Fully documented in text; machine-readable |
| Eastern Ashkenazi Torah mode (scale, resting pitches, cadences) | Precisely documented in text + notation (Malin, Kalib, Portnoy & Wolff, Schmidt chart) |
| The Chabad melody specifically | Not in staff notation in any public source found. Exists as TropeTrainer's Chabad system (sheet music behind demo/subscription) and as Slavin/Schusterman recordings. Reproducing it faithfully needs the TropeTrainer Chabad data or transcription of Slavin by ear/pitch analysis, diffed against the §2.5 chart |
| Rare marks (shalshelet, karnei parah, mercha kefula, yerach ben yomo) | Recordings/forum only |
| Intra-Chabad variation | Real: Slavin's flattened etnachta vs. older two-note etnachta — the tool must name its reference reader |
2.7 Sources for the Chabad melody
- TropeTrainer — Chabad voicing system + sheet music, free demo: https://www.tropetrainer.com/ · https://www.tropetrainer.com/plans ($13/mo or $130/yr; "Chabad" listed under Torah, High Holidays, Haftarah, Esther)
- Schmidt "Torah Trope" chart (transcribed in §2.5): https://www.agudatachim.com/wp-content/uploads/2026/05/torahsheet.pdf
- Malin, "Eastern Ashkenazi Biblical Cantillation" (Yuval 10, 2016), text + notation + audio: https://jewish-music.huji.ac.il/en/yuval/22542
- Jacobson, Chanting the Hebrew Bible (2nd ed. 2017) — ch. 6 gives full notation for all six Ashkenazic systems; companion audio site: https://www.chantingthehebrewbible.com
- Chabad.org — Slavin recordings (all 54 portions + haftarot): https://www.chabad.org/multimedia/audio_cdo/aid/983326/jewish/Torah-Reading-Recordings.htm
- Chabad.org — Interactive Torah Trainer (Slavin voice, labelled "Chabad"): https://www.chabad.org/library/howto/trainer_cdo/aid/2747444/jewish/Interactive-Torah-Trainer-Demo.htm
- Chabad.org — Torah Reading Trop + forum (shalshelet/karnei parah/Slavin etnachta): https://www.chabad.org/multimedia/music_cdo/aid/931078/jewish/Torah-Reading-Trop.htm
- Beverly Hills Chabad — Schusterman recordings + Taamei Hamikra section: https://www.beverlyhillschabad.com/torah-reading/TORAH-MENU.HTM
- ChabadInfo weekly Slavin MP3s (parsha/haftorah/Shnayim Mikra): http://chabadinfo.com/news/prepare-for-krias-hatorah-parshas-naso/
- Lipinski, Bar Mitzvah Torah Trainer (Chabad, 2019) — symbol-only chart, QR audio: https://soundcloud.com/themitzvahclub/sets/barmitzvah-torah-trainer-torah-cantillations-taamei-hamikra · book PDF: https://cdnm.heyzine.com/files/uploaded/v2/6c350fda191e1cd1deda45717fc2048d181aa000.pdf
- Chabad.org — "Who Made Up the Way We Sing the Torah?": https://link.chabad.org/library/article_cdo/aid/817346/jewish/Who-Made-Up-the-Way-We-Sing-the-Torah.htm
- Wikipedia, "Hebrew cantillation": https://en.wikipedia.org/wiki/Hebrew_cantillation
- IvritSuite Trope Tutor (staffs = Avery/Binder Reform — NOT Chabad, do not use as Chabad): https://ivritsuite.com/trope_tutor.html
3. Text data: Chumash with taamim as open data
3.1 Recommended stack
- Text: Sefaria API v3 pinned to the version "Tanach with Ta'amei Hamikra" (public domain, WLC-based), or the same text offline from tanach.us (UXLC). Fetch one verse/chapter at a time from Sefaria, or bulk-download from Sefaria-Export.
- Aliyot: Hebcal
@hebcal/leyning—src/aliyot.json(all 54 parshiyot) +src/holiday-readings.json, BSD-2-Clause. - Audio/melody: there is no open word-aligned text↔audio dataset and no open melody-per-trope dataset. Audio exists only as loose per-aliyah MP3s (Vayavinu Bamikra / Wikimedia Commons). The melody layer would have to be built (recorded, or rule-based synthesis from the trop marks).
- Encoding rule #1: store the text exactly as received — do not NFC-normalize (it corrupts WLC text). Strip marks by code-point range only when toggling nikud/taamim off.
3.2 Sefaria — yes, taamim included
Docs: https://developers.sefaria.org/reference/getting-started — "all of our currently documented endpoints can be accessed without API keys, tokens, or authorization."
Endpoint (v3, current): GET https://www.sefaria.org/api/v3/texts/{Ref} — docs at https://developers.sefaria.org/reference-link/get-v3-texts
Version selection (verified from the docs page): the version param takes language or language|versionTitle. version=hebrew|Tanach with Ta'amei Hamikra returns only that version; no version param defaults to version=primary. There is also an older v2 endpoint https://www.sefaria.org/api/texts/{Ref}.
Verified live call (fetched 2026-11-17):
https://www.sefaria.org/api/v3/texts/Genesis.1.1?version=hebrew|Tanach%20with%20Ta'amei%20Hamikra
Returned (abridged — the versions[0] object only):
{"versions": [{"status": "locked", "priority": 1.0, "license": "Public Domain",
"versionTitle": "Tanach with Ta'amei Hamikra",
"versionTitleInHebrew": "תנ״ך מלווה בטעמי מקרא",
"versionSource": "http://www.tanach.us/Tanach.xml",
"language": "he", "isSource": true, "isPrimary": true, "direction": "rtl",
"text": "בְּרֵאשִׁ֖ית בָּרָ֣א אֱלֹהִ֑ים אֵ֥ת הַשָּׁמַ֖יִם וְאֵ֥ת הָאָֽרֶץ׃"}],
"available_versions": [ ... ]}
So Gen 1:1 with taamim is exactly: בְּרֵאשִׁ֖ית בָּרָ֣א אֱלֹהִ֑ים אֵ֥ת הַשָּׁמַ֖יִם וְאֵ֥ת הָאָֽרֶץ׃ — tipcha ֖ (U+0596) on בְּרֵאשִׁית, merkha ֣ (U+05A5) on בָּרָא, etnahta ֑ (U+0591) on אֱלֹהִים, merkha on אֵת, tipcha on הַשָּׁמַיִם, merkha on וְאֵת, siluk + sof pasuq ׃ (U+05C3) on הָאָרֶץ.
Notes:
- The v2 call https://www.sefaria.org/api/texts/Genesis.1.1 also returned Hebrew with taamim in its he array — but it returned the whole chapter (31 verses) for a verse ref, and mixes in English versions. Use v3 with the pinned version.
- return_format=text_only strips HTML/footnotes; default returns text as stored. (From the v3 docs page.)
- Two Hebrew versions matter, and they differ:
- "Tanach with Ta'amei Hamikra" — Public Domain, sourced from tanach.us/Tanach.xml (WLC). This is the one to use.
- "Miqra according to the Masorah" (MAM) — CC-BY-SA, based on the Aleppo Codex; it is the primary Hebrew version, and it adds editorial accents (e.g. an extra pashta/zarqa/segol/telisha when an accent falls on an unstressed syllable; legarmeh/paseq visibly distinguished; qamatz qatan marked). For leining practice this is a real difference — pick one base text and stick to it.
- A small pointing difference between editions was observed in live fetches: Gen 1:2 Sefaria/MAM has וְהָאָ֗רֶץ הָיְתָ֥ה while Mechon Mamre has וְהָאָ֗רֶץ הָֽיְתָ֥ה. Editions are not byte-identical.
- Licensing: per-version license field in the API response. The Ta'amei Hamikra version is Public Domain; other Sefaria texts vary (CC-BY, CC-BY-NC, copyrighted). Sefaria's links dataset is CC-BY. Sefaria name/logo are trademarked. Source: https://developers.sefaria.org/docs/usage-of-our-name-and-logo
- Bulk download (best for offline): https://github.com/Sefaria/Sefaria-Export — the repo is only an index (~25 MB); the actual data (~26 GB, ~85K files) lives in a public GCS bucket, no auth needed. Paths like gs://sefaria-export/json/Tanakh/Torah/Genesis/Hebrew/Tanach with Ta'amei Hamikra.json, .../txt/.../*.txt; books.json in the repo indexes everything with json_url/txt_url. (The exact Hebrew version filename should be confirmed from books.json before hardcoding it.)
- Prior art: orelzion/2mikra builds a per-aliyah reader on exactly this API (3 parallel calls per aliyah, target versions "Miqra according to the Masorah" or "Tanach with Ta'amei Hamikra") — https://github.com/orelzion/2mikra/blob/HEAD/.agents/skills/sefaria-api/SKILL.md
- Not verified: Sefaria API rate limits / fair-use thresholds. The old API wiki says "If you want to run your own database, please don't use this API to grab our data; instead download a complete dump" — for a shipped app, bulk-download once rather than scraping the API.
3.3 Mechon Mamre — full Tanakh with taamim, HTML only, restrictive licence
- Index: https://www.mechon-mamre.org/c/ct/c0.htm (lists all books and all 54 parshiyot)
- Example chapter (verified live): Genesis 1 with cantillation is https://www.mechon-mamre.org/c/ct/c0101.htm — plain HTML, verse numbers in Hebrew letters, text in Unicode with nikud + taamim. URL pattern is
/c/ct/c{book}{chapter}.htm(book 01 = Genesis). The/c/ct/tree is the cantillated version. - Bulk download: https://www.mechon-mamre.org/dlct.htm — page title "Downloading Bible in Hebrew with Cantillation Marks", single "Click here" link. The exact zip filename/URL was not followed, so it is unverified — flag before relying on it.
- Base text: Aleppo Codex and related Tiberian manuscripts, edited per Rabbi Mordechai Breuer (not the Leningrad Codex, so it differs slightly from Sefaria's WLC version — see the Gen 1:2 example above).
- Format/machine-usability: HTML pages only — no API, no JSON/XML. A scraper would have to parse the HTML. Doable but fragile; Sefaria/tanach.us are better machine sources.
- Licence — the catch: texts are free to read, but reuse is restricted to private study/teaching plus short quotations; republishing needs written permission. Do not use Mechon Mamre as the redistributed text source for a shipped app without permission — Sefaria's PD version or tanach.us is the clean route. Mamre is still useful as a cross-check and for its "Tikkun Korim" (reader's correction) pages.
3.4 Other open sources
tanach.us — Unicode/XML Leningrad Codex (UXLC) ⭐ best offline source
- https://www.tanach.us/ (About: https://www.tanach.us/Pages/About.html, Technical: https://www.tanach.us/Pages/Technical.html)
- A fork of the Westminster Leningrad Codex (WLC 4.20), converted to Unicode/XML, updated twice a year; current build UXLC 2.6 / build 27.7 (18 Oct 2026). Per-book XML files in a
Booksdirectory, validated byTanach.xsd. - Downloads:
Tanach.xml.zip(~2.4 MB), plusTanach.txt.zip,.odt,.pdf. (Direct zip URL not captured verbatim — get it from the Technical page.) - Licence: "You may view or copy all biblical Hebrew text, in any format, without restriction." — the cleanest licence of all the sources.
- This is the same text Sefaria's "Tanach with Ta'amei Hamikra" version is sourced from.
Open Scriptures Hebrew Bible (OSHB / morphhb)
- https://github.com/openscriptures/morphhb (256 stars, created 2009)
- OSIS XML, one file per book in
wlc/, word-level: each<w>has the pointed word (vowels + taamim inline as Unicode) pluslemmaandmorphattributes and a stable per-wordid. Also ships astructure/demo "for displaying verses divided by the cantillation hierarchy" usingnattributes recording cantillation divisions — i.e. the syntactic hierarchy implied by the trop is partially encoded, not just the marks. - Licence: WLC text is Public Domain; lemma/morphology markup is CC-BY 4.0. README explicitly warns: avoid NFC normalization. JSON conversion scripts included. Also contains
MAPM/(Miqra According to the Masorah, regenerated Nov 2026). - Good fit if the tool wants per-word data (e.g. tap-a-word → its trope name).
BHSA / ETCBC (Biblia Hebraica Stuttgartensia Amstelodamensis)
- https://github.com/ETCBC/bhsa — Text-Fabric (
.tf) dataset. Licence CC BY-NC 4.0 (non-commercial — a constraint for a commercial app). - Cantillation is present: the word feature
g_word_utf8contains consonants, vowel pointing and accents inline, and there is a dedicated docs page on cantillation caveats: https://etcbc.github.io/bhsa/cantillation/ — notably the paseq (׀ U+05C0) lives in the separatetrailerfeature, not in the word, and it can change how the preceding accent is read (legarmeh: munach + paseq; a query finds 1,348 candidates, 254 narrow ones). Also: "The BHSA is not committed to maintain a definite ordering" of diacritics within a word. - Richest linguistic annotation (morphology, syntax trees via SHEBANQ), but the Text-Fabric format is heavier to adopt than Sefaria JSON or UXLC XML, and the NC licence bites.
Others (verified as existing, lighter verification)
- STEPBible TAHOT, https://github.com/STEPBible/STEPBible-Data — WLC-derived, CC BY 4.0; Hebrew words keep pointing and cantillation.
- thiagobodruk/bible, https://github.com/thiagobodruk/bible — 141 versions as JSON/XML;
he_wlc= WLC "with vowel points and cantillation marks", MIT-licensed code. Caution: README says files were built by a Python crawler and "may contain minor issues related to encoding and syntax" — prefer the upstream sources above. - hebrew-text-in-practice, https://github.com/ohavben/hebrew-text-in-practice — a pitfalls guide (Oct 2026) built on WLC + Sefaria text: final forms, NFC and mark order, bidi, fonts, invisible characters.
- SWORD / CrossWire OSHB module — same OSHB text in another packaging.
3.5 How taamim are encoded in Unicode
- Range: U+0591–U+05AF (Hebrew accents), plus related points: vowels U+05B0–U+05BB, dagesh/mapiq U+05BC, meteg U+05BD, rafe U+05BF, shin/sin dots U+05C1/U+05C2, maqaf ־ U+05BE (word-joiner, a real character, not a hyphen), paseq ׀ U+05C0, sof pasuq ׃ U+05C3 (verse end), upper dot U+05C4, qamats qatan U+05C7. U+05A2 (atnah hafukh) was added in Unicode 4.1. All are combining marks (general category Mn) that follow the base letter. Source table: http://www.i18nguy.com/unicode/hebrew.html
- Combining classes / canonical order: most accents are class 230 (above), the below-letter accents (etnahta U+0591, tipeha U+0596, tevir U+059B, munach U+05A3, merkha kefula U+05A6, etc.) are class 220; vowels are classes 10–20, meteg 22. Canonical order in stored text is: consonant – dagesh – vowel – low marks – prepositive marks – high marks – postpositive marks. Practically: in Sefaria/WLC text the accent usually comes after the vowel on the same letter, e.g. בְּרֵאשִׁ֖ית = bet+sheva, resh+tsere, alef, shin+hiriq+tipcha U+0596, yod, tav.
- Multiple marks per word are real: a word can carry two accents on one syllable (e.g. Gen 5:29 word 6 and Zeph 2:15 word 1 carry gereshayim + telisha-gedola) and two vowels on one consonant — per the UXLC build-25.1 correction notes. A parser must handle stacks, not assume one mark per letter.
- Normalization gotchas (the big ones):
1. Do not NFC-normalize stored text. The OSHB README warns against it, and
hebrew-text-in-practicemeasured it: in the WLC, NFC "changes almost half the words of the Torah and swaps the two vowels under the lamed of Jerusalem." Store and compare text exactly as received. 2. Do not strip marks by the naive range U+0591–U+05C7 — four code points in that range are punctuation, and three (maqaf U+05BE, paseq U+05C0, sof pasuq U+05C3) separate words/verses rather than decorating a letter. Strip vowels (U+05B0–U+05BB, U+05BC–U+05BD, U+05BF, U+05C1–U+05C2, U+05C7) and accents (U+0591–U+05AF) with separate, explicit sets, and leave maqaf/paseq/sof-pasuq alone if you want to preserve word/verse boundaries. 3. Mark order is not canonicalised by the sources (BHSA explicitly disclaims a stable diacritic order). If you need a canonical key (e.g. for dedup or search), sort combining marks by combining class yourself — don't rely on NFC. 4. Rendering needs a font with OpenType mark-to-mark positioning (GPOSmkmk) covering te'amim as attaching marks over vowels as base marks, and the font must be named first infont-familyor the browser falls back mid-word / draws tofu. Fonts named in the sources: SBL Hebrew, Ezra SIL (used by tanach.us). - Two trop systems: the 21 prose books use one accent system; Psalms, Proverbs, Job use a different one ("Taamei Emet"). For a Torah-only tool this doesn't matter, but don't reuse the Torah melody table for Tehillim.
- Accent ≠ melody, one-to-one: each word normally carries exactly one disjunctive/conjunctive accent, but the musical realisation depends on the accent sequence (phrases), position in the verse, and tradition — the text alone does not spell out notes.
3.6 Aliyot — how divisions are represented
Hebcal @hebcal/leyning (best source — verified in full)
- Repo: https://github.com/hebcal/hebcal-leyning (BSD-2-Clause); npm:
npm install @hebcal/leyning. - Data files (verified live):
src/aliyot.json— all 54 parshiyot. Each entry:num(1=Bereshit … 54=Vezot Haberakhah),book(1–5),haft(+seph,chabadvariants),fullkriyah(keys"1"–"7"plus"M"maftir),weekday(keys"1"–"3", Mon/Thu). Values are[begin, end]verse refs within the book, e.g. Bereshit:"1":["1:1","2:3"], "2":["2:4","2:19"], … "7":["5:25","6:8"], "M":["6:5","6:8"]. Some entries carry a variant note as a 3rd array element, e.g. Bereshit"5":["4:19","4:22","some sources use 4:19-26"]— aliyah boundaries are not 100% universal; the file records the disagreements.src/holiday-readings.json— festival/holiday readings (same shape; keys are holiday names like "Sukkot I").- API (JS):
getLeyningForParsha(parsha),getLeyningForParshaHaShavua(ev, il),getLeyningForHoliday(ev, il),getLeyningForHolidayKey(key, cholHaMoedDay),getLeyningOnDate(hdate, il)— theilflag switches Israel/Diaspora;formatAliyahWithBook(a)renders e.g."Numbers 28:9-28:15"; each aliyah object is{k, b, e, v?, p?}.getLeyningForParshaHaShavuaresolves combined parshiyot and special maftir/haftarah for that Shabbat, and thereasonfield explains special readings. - REST:
https://www.hebcal.com/leyning?cfg=json&start=YYYY-MM-DD&end=YYYY-MM-DD(max 180 days),i=onfor Israel; each event carries aleyningobject withfullkriyah,haftarah,maftir,triennial. - Triennial readings moved to a separate package,
@hebcal/triennial.
Vezot Habracha (verified in the file)
Entry "Vezot Haberakhah": {"num":54, "book":5, … "fullkriyah": {"1":["33:1","33:7"], … "7":["34:1","34:12"]}}. It is read on Simchat Torah (not Shabbat). Notable: unlike every other parsha in the file, it has no "M" (maftir) key — the maftir on Simchat Torah is a repeat of the 7th aliyah (Deut 34:1–12), but that halachic detail was not verified against a source in this pass, so treat the missing M as a data quirk to handle, not as a claim about the reading.
Special cases to handle
- Combined parshiyot (e.g. Vayakhel–Pekudei, Tazria–Metzora, Nitzavim–Vayeilech in some years): Hebcal resolves these by date via
getLeyningForParshaHaShavua— don't hardcode 54 fixed Shabbat readings. - Split/altered aliyot on holidays: holiday readings have 3–6 aliyot, not 7 (Yom Kippur 6, festivals 5, Chol HaMoed/Rosh Chodesh 4, fast days 3); Shabbat that coincides with Rosh Chodesh / Chanukah / a festival replaces the maftir (and sometimes the 7th aliyah) — Hebcal encodes this via
getLeyningForHoliday+ thereasonmap. - Sefaria also carries aliyah data: Sefaria's Calendars API returns parasha objects whose
extraDetailscontains analiyotstring array (seen in theryanfaerman/go-sefariapackage docs). Not live-verified — Hebcal is the verified source; Sefaria's is a cross-check. - Weekday (Mon/Thu) readings: included in Hebcal as
weekday(first 3 aliyot, shorter ranges) and in Sefaria via the same calendar data.
3.7 What is NOT available as open data
- No open word-aligned text↔audio dataset. Searches for a Torah cantillation audio dataset / word-aligned leining corpus found only academic MIR papers (UVic: Ness/Tzanetakis/Biro on melodic-contour analysis of Torah trope; Technion CIM14: automatic analysis of Yemenite cantillation, corpus = two readers recorded at Hebrew University's Jewish Oral Traditions Research Center). These are research corpora, not released open datasets — no download link, no timestamps, no licence found.
- Audio exists, but unstructured: the Vayavinu Bamikra project (Wikisource, https://en.wikisource.org/wiki/Vayavinu_Bamikra) organises free recordings on Wikimedia Commons (categories "Cantillation" / "Jewish cantillation") — 400+ recordings of aliyot, haftarot, megillot, per-parsha pages per book. They are whole-aliyah MP3s with no word timings, no structured index verified, and mixed readers/traditions. Usable as reference listening; not usable as training/alignment data without manual segmentation.
- No open melody dataset: the mapping "trope name → melody" is oral tradition, varies by community, by book, and by position in the phrase. There is no standard open file (MIDI, MusicXML, pitch track) encoding e.g. "Chabad Torah trop". Chabad-specific trop melody was not verified anywhere in this pass — do not assume it equals generic Ashkenazi; that needs a cantor/musicologist source or fresh recordings.
- No open "trope name per word" layer as a separate dataset — but it is derivable: each word in the sources above already carries its accent(s) as Unicode, so the per-word trope label is a parse of the text, not a separate lookup. Edge cases: two accents on one word; legarmeh (munach + paseq, needs the
trailer); MAM's added accents vs WLC. - Commercial/closed alternatives (not open data, for awareness): TropeTrainer, Leining Master (Rabbi Jeremy Wieder — full Torah/Haftarah/Megillot MP3s, free individual downloads/streaming at YU Torah), PocketTorah, Navigating the Bible. These solve the audio problem by recording, not by an open dataset.
3.8 Suggested data architecture (one paragraph)
Ship the text offline: bulk-download Sefaria-Export's Tanach with Ta'amei Hamikra (PD) or the UXLC Tanach.xml.zip, store verses as received (no NFC), index by book/chapter/verse. Drive the practice flow from Hebcal @hebcal/leyning (aliyot.json + holiday-readings.json, or the REST API for date-aware readings incl. Israel/Diaspora). Derive per-word trope labels by parsing the Unicode accents in each word (handle stacks, maqaf-joined words, paseq/legarmeh). For audio, plan on recording your own reference leining per aliyah (or licensing Wieder/YU Torah streams) and aligning it yourself — that alignment layer is the genuinely missing open data, and it's the thing this tool would have to create.
4. What already exists — and the gaps
4.1 Chabad.org Torah Trainer (+ Megillah Trainer) — the only Chabad-native tool
- What it does: Interactive web trainer for all 54 Torah portions, all haftarahs, and the blessings before/after readings. A sister tool, the Megillah Trainer, covers Megillat Esther. It guides students word by word through the portion.
- Platform: Web (desktop + mobile browsers); no native app. Live at chabad.org/torahtrainer.
- Price: Free — no price or paywall found anywhere; Chabad.org tools are free.
- Voice: Recorded human voices, three of them: Rabbi Chayim B. Alevsky (Modern Hebrew intonation), Rabbi Levi Kaplan (traditional Russian-style Eastern European, baritone; director of the Council of Chabad Cantors), and Rabbi Michoel Slavin (added Feb 2025 to both trainers). This is the Chabad trop tradition.
- Practice features: Word-by-word highlighting as the audio plays; tap any word/phrase to replay it; playback speeds 0.5×, 1×, 1.5×, 2×; playback modes Phrase / Verse / Complete; fonts Stam / Standard / Vowels; show/hide cantillation marks, transliteration, translation.
- Limitations: It is essentially a smart recording player, not a teacher. No record-yourself, no pitch/melody feedback, no quizzes or drills, no trope-system lessons (it never teaches why a mark sounds the way it does), no progress tracking, no offline use, no speed below 0.5×, and no way to loop a single hard phrase automatically.
- Sources: https://www.chabad.org/blogs/blog_cdo/aid/3352660/jewish/The-Ultimate-Torah-Reading-App.htm · https://www.chabad.org/library/howto/trainer_cdo/aid/2747444/jewish/Interactive-Torah-Trainer-Demo.htm · https://anash.org/leining-megillah-just-got-that-much-easier/
4.2 TropeTrainer (HazzanSolutions, successor to Kinnor Software)
- What it does: The most comprehensive trope trainer; full Tanach (all 54 portions + every aliyah, all haftarahs, Five Megillot, High Holidays, special Shabbatot, Rosh Chodesh, fast days, weekday readings) in multiple text views.
- Platform: Web app, any browser/phone/tablet/computer; no install. (An older Kinnor mobile app also exists on iOS with per-reading in-app purchases, $7.99/reading, $59.99 all Torah+Haftarah, $79.99 all incl. Megillot — but the current web product is a separate subscription and old purchases do not carry over.)
- Price: $13/month or $130/year; 30-day money-back guarantee; no free trial, but a full demo reading is playable without an account. Organization seats up to 35% off.
- Voice: 12 voice styles, speeds 0.25×–2×, 6 accent traditions, word highlighting as sung. Unverified whether the web version's voices are recorded humans or synthesized — the old Kinnor mobile app explicitly used DECtalk text-to-speech synthesis; the new site only says "voice styles." Flag this before relying on it. Wikipedia lists TropeTrainer under "Mechanical Cantillation (computer speech synthesis)".
- Traditions: 79+ voicing systems including Chabad, Avery/Binder, J. Jacobson, Rosowsky/JTS, Spiro, and Sephardi traditions — the only non-Chabad tool that offers a Chabad system at all.
- Practice features: Guided trope tutor lessons; Vowels & Trope / Torah STaM / Tikkun (side-by-side) views; customizable saved Practice Views; trope-phrase color coding with sheet music for every melody; record your voice in-browser and download it to share with a tutor; print/PDF export; perpetual Hebrew/Gregorian calendar.
- Limitations: Paid subscription is the main barrier; requires internet (offline only via printed PDFs); no pitch feedback on your recording (record-and-compare only); no quizzes/spaced repetition; Chabad is one system among 79, not the default or the focus; no progress tracking or gamification.
- Sources: https://www.tropetrainer.com/ · https://www.tropetrainer.com/plans · https://www.tropetrainer.com/help
4.3 Pocket Torah (+ PocketTorah Trope) — Russel Neiss
- What it does: Free app for learning the weekly Torah and Haftarah; tap any verse number to hear it chanted with text highlighted; Tikkun View toggles between vowelled text and scroll (STaM) script. Companion app PocketTorah Trope teaches just the trope melodies with tutorials and practice exercises. Audio is also embedded in Sefaria's Torah text pages.
- Platform: iOS and Android; also a web version at pockettorah.com.
- Price: Free, no in-app purchases or subscriptions.
- Voice: Recorded human voices (multiple cantors across readings; the Trope app's chanter is a woman — see complaints below). One trope system only, essentially a Reform/Avery-Binder-flavoured Ashkenazi melody.
- Practice features: On-demand audio per verse, tap a word to repeat it, whole-aliyah continuous play with play/pause, text highlighting synced to audio, Tikkun view, English translation.
- Limitations / user complaints (from App Store reviews):
- Only one trope version — "the trope vocalizations I learned and teach are not represented here. I won't be able to use this program"; the trope system is not even consistent across all weekly readings.
- Bugs: highlighting drifts out of sync with the chanting; pause button often doesn't work or is delayed; app lags/crashes, forcing a restart every 4–5 sections and re-finding your portion; no search or save-your-portion feature.
- Content gaps: some special haftarahs missing (e.g. Zachor); at least one haftarah where the chanting doesn't match the text (Vaera — Kings 4:1–37).
- Kol isha: a woman chants in PocketTorah Trope, which some Orthodox users object to ("Not suitable for religious users").
- Stale: last major version 4.0.0 dates to 2021; PocketTorah Trope missing rare tropes (Shalshelet, Mercha Kefulah, Karnei Farah, Yerach Ben Yomo).
- No slow playback, no record-yourself, no quizzes, no pitch feedback.
- Sources: https://apprecs.com/ios/497873394/pockettorah-trope · https://apptail.io/app/pockettorah-xhef/pakistan
4.4 Trope on the Go (Studio HaMikra, cantor-built, launched ~Sept 2026)
- What it does: The newest and most pedagogy-driven tool: bottom-up method — learn one trope at a time, map it to syllables, practise real words, then read your portion.
- Platform: Web app (any device) + iPhone app (iOS 17+); native Android planned, not yet released.
- Price: Free tier (learn every trope sound + sequence, voice setup, Trope Catcher game, one full Torah aliyah or Haftarah with standard tikkun) after a 7-day full-access trial, no card required. Core: $11.99/month or $119.99/year. Clergy/educators/tutors may qualify for free lifetime access.
- Voice: Recorded professional cantor — explicitly "hear it sung, not synthesized." Voice Setup calibrates playback to your comfortable chanting key. Ashkenazi only; Sephardi/Mizrahi and High Holiday/Festival melodies are "on the roadmap." No Chabad tradition.
- Practice features (the strongest set in the field): Animated "sound maps" showing how a melody lands on each syllable of any word; real-time pitch detection with a pitch trail vs. target notes; record & compare; 4,000+ common word/phrase library; focus filters (by trope, syllable count, accent position); 42-line tikkun matching modern scrolls plus enhanced tikkun with transliteration/hyphenation/spacing controls and a trope-transliteration view for learners who can't read Hebrew yet; flashcards with spaced repetition; games (Trope Catcher, dictation, matching, milra/mileil sorting, sequencing); speed control; crash course.
- Limitations: No Chabad or Sephardi system yet; no High Holiday/Festival trope yet; iOS-only native app (web works everywhere); very new (v1.1, Oct 13, 2026) so long-term reliability is unproven; subscription paywalls unlimited readings.
- Sources: https://tropeonthego.com/splash · https://apps.apple.com/us/app/trope-on-the-go-chant-torah/id6802778249
4.5 LearnTrope.com
- What it does: Free interactive lessons teaching Torah and Haftarah trope as seven common melodic phrase-groups, with visual marks, written musical notation, downloadable MP3s, hand-signal chart, and guided practice (each Shema line has three audio files: spoken Hebrew, trope alone, both combined).
- Platform: Web only. Price: Free (donation-supported).
- Voice: Recorded audio (MP3s). Tradition is its own system — teachers warn it uses a slightly different Torah system than many congregations teach, and "completely different Haftarah trope melodies," so some teachers don't recommend it for Haftarah.
- Limitations: Teaches the trope system, not your portion — no full readings, no tikkun, no record/compare, no quizzes, dated site design.
- Sources: https://www.learntrope.com/ · http://learntrope.com/practice.html
4.6 TorahTrope.com
- What it does: Free web reader/trainer: all 54 portions + High Holiday readings in a scholarly Masoretic text, with recorded Ashkenazi audio for every aliyah, slowable to half speed, color-coded trope groupings, tap any word to see its trope name, custom verse ranges (e.g. Deut 21:10–21) with audio trimmed to match and downloadable clips, three display modes (full text, vowels only, consonants only), printable pages. No account.
- Platform: Web. Price: Free.
- Limitations: Ashkenazi audio only, no Chabad; no record-yourself, no pitch feedback, no quizzes/progress tracking; a practice reader, not a full tutor.
- Sources: https://torahtrope.com/pdf/how-to-practice-torah-trope.pdf
4.7 IvritSuite — Trope Tutor + Torah Trainer (free web)
- What it does: Trope Tutor teaches the ~26 marks by clause family (symbol, Ashkenazi + Sephardi names, melody on a staff) with real chanted examples, then Drill mode quizzes three ways (see mark → name it; hear clip → pick symbol; see symbol → pick melody) with per-mark mastery tracking — missed marks recur more often, and distractors get sneakier (e.g. pashta vs. kadma) as mastery grows. Torah Trainer is a companion reader with vowel coloring, transliteration, translation, cantillation toggle, and chanted-audio karaoke. Free, no account, no ads.
- Voice: Reuses PocketTorah's recordings (Reform/Avery-Binder melody; CC BY-SA 4.0) plus Cantor Arianne Brown's Conservative recordings by permission.
- Limitations: Reform/Conservative melodies only — no Chabad; Trope Tutor is a drill tool, not a portion-prep tool; web only.
- Sources: https://ivritsuite.com/trope_tutor.html · https://ivritsuite.com/
4.8 Torah Morah (Adam Bellows)
- What it does: iPhone app teaching Hebrew decoding and Torah trope through guided lessons ("Trope Path" — each trope to its trope family), interactive quizzes and games, and "Chant It!" — sing along and get note-by-note pitch feedback. Also handwriting practice with feedback and word-by-word pronunciation feedback; full Torah text with recordings of every aliyah, downloadable for offline use (via Sefaria).
- Platform: iPhone only. Price: Free to download (could not verify whether lessons beyond the intro are paid — no pricing found on its site or store copy read).
- Voice/tradition: Recordings; tradition not specified on the site read — unverified, likely standard Ashkenazi.
- Limitations: iPhone only; no Chabad system advertised; pricing beyond "free to download" unclear.
- Sources: https://torahmorah.org/
4.9 Pink Trope (Danielius Studio, new Oct–Nov 2026)
- What it does: Free iPhone app, tagline "Chant it. It listens." Find your portion by week or passage; pointed Hebrew text; every trope mark shown as its tune (Ashkenazi, with a Sephardi list also shown); it listens to you chant and scores each attempt (separate Words % and Melody %); an Archive keeps and compares every attempt.
- Platform: iOS only. Price: Free on the App Store (1 rating so far — brand new).
- Limitations: Ashkenazi only, no Chabad; brand-new and unproven (accuracy of its listening/scoring unverified); no pedagogy content found — it appears to be a scorer, not a teacher.
- Sources: https://www.pinkfame.art/
4.10 Also noted (not deep-dived)
- Sefaria embeds PocketTorah audio verse-by-verse — a free reading aid, not a trainer. https://www.sefaria.org/texts
- Hello Haftarah — karaoke-style Torah portions for bar/bat mitzvah prep (mentioned in search results; not independently verified). https://www.hellohaftarah.com/getting-your-teen-ready-for-their-bar-bat-mitzvah-memorize-torah-portion/
- Navigating the Bible (ORT) — free recorded parashot/haftarot; older site, Ashkenazi. http://bible.ort.org/
- Tikkun Korim — the user's own app (iOS/Android tikkun reader); a tikkun/text tool, not a trope tutor.
4.11 Comparison table
| Tool | Platform | Price | Voice | Chabad trop? | Follow-along highlight | Slow playback | Repeat a pasuk/word | Record yourself | Pitch feedback | Quizzes/drills |
|---|---|---|---|---|---|---|---|---|---|---|
| Chabad Torah Trainer | Web | Free | Recorded human (Alevsky / Kaplan / Slavin) | Yes — native | Yes, word-by-word | 0.5×–2× | Tap word/phrase | No | No | No |
| TropeTrainer | Web (+old iOS) | $13/mo, $130/yr | 12 "voice styles" — recorded vs. synth unverified for web version | Yes, 1 of 79 systems | Yes | 0.25×–2× | By word/verse/aliyah | Yes (record, playback, download) | No | Tutor lessons only |
| Pocket Torah | iOS/Android/web | Free | Recorded human (multiple cantors) | No | Yes | No | Tap verse/word | No | No | No (Trope app has tutorials/exercises) |
| Trope on the Go | Web + iOS | Free tier; $11.99/mo, $119.99/yr | Recorded cantor, not synthesized | No (Ashkenazi only) | Yes | Yes | Word/phrase repeat | Yes, record & compare | Yes, real-time pitch trail | Yes — flashcards (spaced repetition) + games |
| LearnTrope | Web | Free | Recorded MP3s | No | No | No | Line-by-line audio | No | No | Guided practice only |
| TorahTrope.com | Web | Free | Recorded, Ashkenazi | No | Tap word → trope name | Half speed | Custom verse ranges | No | No | No |
| IvritSuite Trope Tutor | Web | Free | PocketTorah recordings (Reform) | No | No | No | Clip replay in drills | No | No | Yes — 3 quiz modes, mastery tracking |
| Torah Morah | iPhone | Free download (rest unverified) | Recorded | Not advertised | Not verified | Not verified | Not verified | Yes (sing-along) | Yes, note-by-note | Yes — quizzes/games |
| Pink Trope | iOS | Free | Male voice in demos | No | Not verified | Not verified | Not verified | Yes (it listens) | Yes — Words % + Melody % scores | No (archive/scores only) |
4.12 Gaps a new Chabad lein-practice tool could fill
- A Chabad-trop tutor, not just a player. Chabad.org's trainer is a playback tool; TropeTrainer offers Chabad as 1 of 79 systems with no Chabad-first pedagogy. Nobody teaches the Chabad system clause-by-clause (families, frequency order, lookalike drills for kadma/pashta and yetiv/mahpach in the Chabad melody).
- Record-yourself + pitch feedback in the Chabad melody. Trope on the Go, Torah Morah, and Pink Trope all do pitch/score feedback — but only in Ashkenazi (non-Chabad) melodies. A Chabad-melody version of this does not exist.
- A male voice, by default. PocketTorah Trope's female voice draws kol-isha complaints from Orthodox users; a Chabad tool with a male baal koreh voice (e.g. in the style of Alevsky/Kaplan/Slavin) removes that objection entirely.
- Phrase-level practice loop for an aliyah. Tap-to-repeat a word exists in Chabad's trainer, but nobody offers phrase looping, sub-0.5× slow playback, "loop my hard spots," or a structured daily session (15–20 min) built around one aliyah.
- Lookalike-pair drills in Chabad trop. The known confusion pairs are a proven drill format (IvritSuite, Brainscape) — none exists for the Chabad melody, where the pairs sound different than in Avery/Binder.
- Chabad-specific content coverage. Weekday (Mon/Thu) readings, Rosh Chodesh, fast days, holidays, and Megillah in Chabad trop, plus the aliyah blessings — Chabad's trainer has the texts, but no practice structure around them; TropeTrainer has the structure, but Chabad is an afterthought.
- Progress tracking + teacher sharing. TropeTrainer lets you download a recording to send a tutor; nothing offers per-pasuk mastery tracking, spaced-repetition review of weak phrases, or a tutor dashboard for a Chabad cheder/shul.
- Free, offline-capable, Hebrew-first. The free tools (Pocket Torah, TorahTrope, IvritSuite) are Ashkenazi/Reform and buggy or stale (Pocket Torah's sync/pause bugs, 2021 last major update); the good pedagogy tools cost $12–13/month. A free, maintained, offline-capable Chabad tool would be the first in its lane.
- Trust/accuracy gap. Pocket Torah has documented wrong-text/wrong-audio mismatches and missing special haftarahs; a Chabad tool that is verified against the Chabad text (and clearly states its source) would be a differentiator.
Bottom line: the field splits into (a) free recording players with the wrong melody and real bugs, (b) one excellent but expensive generalist (TropeTrainer) that includes Chabad as an option, (c) one excellent new pedagogy app (Trope on the Go) that is Ashkenazi-only, and (d) Chabad.org's own trainer, which is free, authentic, and Chabad — but only a player. The open space is exactly a Chabad-first teacher: Chabad melody, male voice, clause-by-clause lessons, phrase looping, lookalike drills, record + pitch feedback, tikkun-to-STaM progression, and progress tracking.
5. How people actually learn to lein (pedagogy)
5.1 The traditional method
- Trope = short musical phrases, not notes. There are ~28–30 marks; each stands for a short melodic phrase, and the marks simultaneously carry melody, mark the stressed syllable, and punctuate (like commas/semicolons/full stops). The grammar/punctuation job is the one beginners most often miss.
- Learn phrases/clauses, not single marks and not whole verses. The standard teaching sequence (taught by cantors and in courses such as Tamar Kamionkowski's "Learn to Leyn"): introduce one trope clause (a family of marks that occur together) at a time; drill symbol → name → tune until the association is automatic (like hearing "Twinkle Twinkle" when you read the words); then attach real Torah words; then integrate clauses until you can chant a whole verse. Trope on the Go explicitly copies this "part-to-whole" pedagogy; the older "Royals and Servants" classroom method does the same with one "Royal" (disjunctive) mark per lesson.
- Teacher vs. self-study: Teachers are the norm for bar/bat mitzvah (weekly tutor + daily home practice). Self-study works for adults who already read Hebrew, but every guide stresses checking your tradition with a cantor/tutor — the marks are identical everywhere, but the melodies differ by community, and learning the wrong melody is the classic self-study failure. One teacher's advice: practice each new trope group for at least a week before moving on; use flashcards until symbol/name/melody recognition is automatic.
- Sources: https://torahtrope.com/pdf/how-to-practice-torah-trope.pdf · https://kamionkowski-bet-midrash.teachable.com/p/learn-to-leyn · https://raeandesigns.wixsite.com/hebrewresources/trope-introduction · https://www.torahreadingguide.com/beginners
5.2 A good practice loop
The TorahTrope guide's six-step method, corroborated by the Torah Reading Guide:
- Read the words first, without chanting — get Hebrew fluent as plain reading; decoding letters and holding a melody at once means neither sticks.
- Find the phrase boundaries first — etnachta (mid-verse) and sof pasuk (verse end) divide each verse into halves/quarters; chanting is easier when you know where you're heading.
- Learn one phrase at a time, not one verse — chant the words up to the next separating mark until automatic, then add the next group. Verses are too long to hold at once.
- Loop the hard parts — the 2–3 spots that keep going wrong deserve most of the time; re-reading a whole aliyah to fix four words wastes the other three minutes.
- Practice out loud, at full volume, standing up — quiet mental practice builds a different skill than chanting in front of people.
- Drop the training wheels early — once solid, practise from consonants only (STaM/tikkun Torah side), leaving enough weeks to do it without panic. Progression: vowels+trope → two-column tikkun → STaM only.
Supporting habits: short frequent sessions beat long rare ones (15–20 min several times a week); record yourself and compare; use a teacher to correct errors before they fossilize.
5.3 How long it takes
- A typical single aliyah: 3–6 months of regular short practice (15–20 minutes, several times a week). A full parashah, or a student also learning haftarah and prayers, takes longer.
- Bar/bat mitzvah overall: most families start 6–12 months ahead; synagogue guidance says 6–18 months depending on Hebrew level, with 9–12 months before the date for tutoring to begin and daily 15–20-minute home practice from ~7–8 months out. Chabad's own bar mitzvah guidance: optimal preparation 6–18 months.
- Learning the whole system (not just one portion) is a longer project — months of clause-by-clause work — but it transfers to any portion, whereas memorizing one portion from a recording does not.
- Sources: https://torahtrope.com/pdf/how-to-practice-torah-trope.pdf · https://www.jewishoakland.org/learn/bar-bat-mitzvah-timeline/ · https://www.hellohaftarah.com/getting-your-teen-ready-for-their-bar-bat-mitzvah-memorize-torah-portion/
5.4 Common stumbling blocks
- Lookalike marks (the #1 confusion): kadma vs. pashta (identical shape; kadma sits over the accented syllable, pashta always over the last letter of the word); yetiv vs. mahpach (identical shape; yetiv sits before the word, mahpach under the accented syllable — rule of thumb: mahpach is always followed by pashta, yetiv never is, with two exceptions in the Torah, Lev 5:2 and Deut 1:4); azla vs. gereish; meteg (not a trope, no sound — secondary stress) vs. siluk; munach legarmei vs. p'sik (pause mark). Drill tools deliberately use these pairs as "sneaky" distractors.
- Same mark, different sound in context: e.g. a mercha before darga sounds different from a mercha before tipcha; munach changes by what follows it. Learners who memorize marks in isolation get tripped in real text.
- Rare tropes: shalshelet, mercha kefulah, karnei farah, yerach ben yomo appear so rarely that learners (and some apps) skip them — then freeze when they appear.
- Memorizing the portion vs. learning to read: the most-cited failure mode — "a reader who can perform their portion and cannot read anything else." Handing someone a recording produces this; phrase-by-phrase system learning avoids it.
- The STaM transition: going from vowelled/cantillated text to bare scroll letters is a distinct skill that needs its own practice time; students who leave it to the last weeks panic.
- Long aliyot / stamina: long readings (and long verses with many conjunctives) punish anyone who practises verse-by-verse instead of phrase-by-phrase; also breathing/phrasing at etnachta and sof pasuk.
- Wrong tradition: learning a Reform (Avery/Binder) melody when your shul uses another, or vice versa — the single most common wasted-effort complaint about free tools.
- Sources: https://api.brainscape.com:443/flashcards/6-lookalikes-and-other-symbols-3026944/packs/4707917 · https://www.davecurtis.net/haph/haphnote.html · https://ivritsuite.com/trope_tutor.html
5.5 What makes a practice tool effective vs. just a recording player
From the guides and from what the best tools do: (1) teach the system (clause families, in frequency order) rather than only playing your portion; (2) phrase-level looping and slow playback, not just whole-aliyah play; (3) visible phrase boundaries (color-coding, stacking by phrase); (4) a tikkun progression toward STaM; (5) active recall — quizzes, flashcards with spaced repetition, and lookalike-pair drills; (6) record-and-compare, ideally with pitch feedback, so errors get caught between lessons; (7) the right tradition's melody, with a male voice for Orthodox users; (8) short-session structure and progress tracking so daily practice actually happens.
6. The technical fork: how the voice gets made
The question that started this research: for producing the voice, compare (a) a tiny TTS with rule-based trop rendering from the marks, and (b) a recorded baal koreh voice. What's actually feasible for each, what data each needs, and what the hard parts are.
6.1 First: no TTS sings taamim today
Nothing found in this pass — no open model, no commercial product, no paper — produces chanted Torah with cantillation from text. Hebrew TTS exists, but for spoken Modern Hebrew, not chanted Biblical Hebrew. The melody layer is missing everywhere. This is the central fact of the fork: whichever path is chosen, the trop melody has to be supplied by the project, not by an off-the-shelf model.
The one precedent for machine-made leining is TropeTrainer, which Wikipedia lists under "Mechanical Cantillation (computer speech synthesis)" — it synthesizes chant with older (pre-neural) speech technology, in multiple melody systems. So synthesized leining is a proven product shape; it just hasn't been done with a tiny neural TTS.
6.2 What exists that each approach can build on
Hebrew TTS (spoken, modern) — the voice half of approach (a): - Phonikud (Interspeech 2026, https://phonikud.github.io, paper https://arxiv.org/abs/2506.12311): open-source Hebrew grapheme-to-phoneme producing fully-specified IPA, driving small real-time TTS like Piper (32M parameters, VITS) trained on 20 hours of synthetic Hebrew audio (via Gemini 2.5 Pro). Fine-tunes TTS with as little as 2 hours of data; runs on edge devices (Raspberry Pi); open models, code, and data (ILSpeech corpus). This is the Hebrew analogue of the tiny-voca recipe: synthetic data + small model. - BlueTTS (MIT, https://github.com/maxmelichov/BlueTTS, https://bluetts.com): multilingual TTS including Hebrew, ONNX, CPU-first, Hebrew G2P via renikud, runs in the browser on WebGPU with a CPU fallback. - HebTTS (Hebrew University, https://github.com/slp-rl/HebTTS, paper https://arxiv.org/pdf/2407.12206): open-source Hebrew TTS from ~5,000 hours of in-the-wild recordings; diacritic-free operation. - Key nuance for Torah text: the input is already vocalized — nikud is the pronunciation, so the G2P problem that Phonikud solves for unvocalized Modern Hebrew is mostly solved by the text itself. What remains is (i) pronunciation tradition: open Hebrew TTS is Modern Israeli pronunciation; Chabad leining needs Ashkenazi pronunciation (different vowels: e.g. tav as "s", kamatz as "aw", cholam as "oy" in some positions), which no open model provides — a voice would need training/fine-tuning on Ashkenazi speech; and (ii) the melody, which no TTS provides at all.
The pitch-control mechanism — the melody half of approach (a):
- tiny-voca voice tuner (Nov 16, 2026, https://muninn.austegard.com/blog/tiny-voca-voice-tuner.html) — verified from the page: tiny-voca accepts per-phoneme duration and per-frame pitch overrides. "Two inputs override the model's own predictions: how many frames each phoneme lasts and the pitch of each frame. Lyrics are written in syllables (A-maz-ing grace) and notes as D4/1, or B4/.5~G4/.5 for a syllable held across two notes. Consonants keep roughly their spoken length and vowels fill the rest of each note." Result: "Told each phoneme's length and pitch the speech model sings within two cents of the notes" — median note error 0.7 to 1.4 cents, every note within 50 cents of the score (Praat pitch tracker). The model is 6.3 MB, runs in the browser, speaks ~27× real time on an iPhone.
- This is the exact control interface a trop renderer needs: syllables + note sequence in, chanted voice out. It proves the architecture class (small neural vocoder with explicit duration + pitch inputs) can sing accurately — it does not prove a Hebrew model can, but the mechanism is language-independent; the phoneme inventory and the voice are what would change.
- Related: Samuel (https://vvolhejn.com/blog/2026-08-17-samuel) — a speech model built on Pink Trombone (a vocal-tract simulator) where changing the pitch contour makes the voice sing a given melody. Speech-to-speech, not TTS, but shows the same principle: pitch is an external, controllable input.
- Research-grade singing synthesis (YingMusic-Singer, Interspeech 2026, https://www.isca-archive.org/interspeech_2026/hao26_interspeech.pdf) does melody-controllable singing with diffusion models — but these are large research models, not tiny on-device ones, and they need a melody-providing reference clip.
Recorded audio — approach (b): - The reference recordings already exist: Slavin (all 54 portions + haftarot, Chabad.org), Schusterman (Beverly Hills Chabad), Wieder/YU Torah (full Torah/Haftarah/Megillot MP3s, free individual downloads). What does not exist is word-aligned audio — per-word timings for any of these. That alignment layer would have to be built (forced alignment of the recording against the known text, or manual segmentation per aliyah). - The whole Torah is finite: 54 parshiyot, ~5,846 verses. A single reader can record the entire Torah once; every aliyah is then covered forever.
6.3 Approach (a): tiny TTS + rule-based trop rendering
| Aspect | Assessment |
|---|---|
| What it is | A small Hebrew TTS voice (Ashkenazi pronunciation) driven by a trop renderer: parse text → phrases → motif table → per-syllable note + duration sequence → pitch/duration overrides into the TTS. The melody is code; the voice is a model. |
| What it needs | (1) A validated Chabad motif table (§2.5, validated against Slavin — the missing piece); (2) a phrase parser (§1.5); (3) a Hebrew TTS voice with Ashkenazi pronunciation — none exists open; needs fine-tuning on Ashkenazi speech (Phonikud's "as little as 2 hours of data" is the encouraging datapoint, but the 2 hours must be Ashkenazi, and ideally chanted, not spoken); (4) a duration model (how long each syllable/note lasts — trop has no fixed meter, so this is a design choice, not a fact to look up); (5) the text + aliyah data (§3). |
| Hard parts | The motif table validation (Chabad melody is recordings-only — §2.6); the Ashkenazi voice (no open source; recording/fine-tuning a voice is a real project); the duration/rhythm model (free chant — what sounds right is a judgment call, needs a cantor's ear to validate); quality risk — a slightly-wrong synthetic chant teaches wrong melody, which is worse than no chant for a learner. |
| Strengths | Infinite coverage: any verse, any aliyah, any custom range, any speed, any pitch/key, on demand, offline, tiny (MBs). One voice, every reading. Follows the tiny-voca spirit: own the voice outright, no recordings of a real person needed (though a real voice is still the best fine-tuning data). Can render the rare tropes (shalshelet etc.) the moment the motif table has them. |
| Feasibility verdict | Plausible, but the hardest path. The pitch-control mechanism is proven (tiny-voca sings within cents); the Hebrew small-TTS recipe is proven (Phonikud); what is unproven is the combination (Hebrew + Ashkenazi + chanted) and the motif table it depends on. Estimate: real R&D, not a weekend. |
6.4 Approach (b): recorded baal koreh voice
| Aspect | Assessment |
|---|---|
| What it is | A real reader (e.g. in the style of Slavin) records the Torah — per aliyah, or per verse — and the tool plays it back with word-level highlighting, phrase looping, and speed control. The melody is in the audio; the tool's job is alignment + pedagogy. |
| What it needs | (1) A reader and recording sessions (or a licence to existing recordings — Slavin's are on Chabad.org, Wieder's at YU Torah; licensing terms unverified); (2) word/phrase-level alignment of audio to text (forced alignment, or manual segmentation — the missing open-data layer, §3.7); (3) the text + aliyah data (§3); (4) the phrase parser (§1.5) for looping/drills (the parser is needed either way). |
| Hard parts | Alignment: getting per-word timings right across ~5,846 verses is the bulk of the work (forced alignment on Hebrew chant is not an off-the-shelf solved problem; manual segmentation is laborious but tractable per aliyah). Recording quality/consistency across sessions. Speed control: time-stretching audio below ~0.5× degrades; a practice tool wants very slow playback. Fixed voice: one reader, one style — but for Chabad practice that is a feature (one canonical reader). |
| Strengths | Authentic from day one — the melody is exactly right, including the rare tropes, the coda, the phrasing, the breathing. No ML risk, no "slightly wrong melody teaches wrong" failure mode. Proven by every existing tool (Chabad.org trainer, Pocket Torah, Trope on the Go all use recordings). |
| Feasibility verdict | The safe, fast path. The work is recording + alignment + app, all well-understood. The main cost is the recording sessions and the alignment effort, not research. |
6.5 A third option worth naming: hybrid
Record the reference voice and use it to bootstrap the synthetic one: the recordings give (i) the ground-truth motif table (pitch-track Slavin, diff against the §2.5 chart — §7 step 1), (ii) the fine-tuning data for an Ashkenazi chanted voice (a few hours of Slavin-style chant is exactly the "2 hours of data" Phonikud-style fine-tuning wants), and (iii) the MVP's audio while the synthetic voice is being built. The recorded path ships first; the synthetic path replaces/augments it later (e.g. for arbitrary verse ranges, speed/key changes, and offline size). This is the recommended sequencing — see §7.
6.6 What would have to be built either way
Both approaches need: the phrase parser (§1.5), the text + aliyah pipeline (§3), the motif table or the alignment layer, and the practice loop (§5). The voice is the only fork. Everything else is shared infrastructure — so the fork can be decided late, after the shared pieces exist.
7. Recommended path
Stage 0 — Validate the melody (the one thing everything depends on). 1. Open the TropeTrainer free demo, select the Chabad voicing system, inspect its sheet music — if its note data is accessible, that becomes the canonical motif table. 2. In parallel, take the §2.5 chart as the working hypothesis. 3. Validate against Slavin's Chabad.org recordings: transcribe by ear/pitch-tracking the first verses of Bereishit (which contain merkha-tipcha-munach-etnachta, katon, sof pasuk) and diff against §2.5. Pay special attention to: etnachta (Slavin reportedly flattens it), the sof-pasuk ending (3–1 vs 1–5), and the shalshelet (2 pazers?). 4. Name the reference reader in the tool (Slavin) and record the decision — intra-Chabad variation is real.
Stage 1 — Shared infrastructure (no voice decision needed).
5. Text pipeline: bulk-download Sefaria-Export's Tanach with Ta'amei Hamikra (PD) or UXLC Tanach.xml.zip; store as received (no NFC); index by book/chapter/verse.
6. Aliyot: Hebcal @hebcal/leyning (aliyot.json + holiday-readings.json).
7. Phrase parser: tokenise by codepoint (handle the §1.5 collisions), parse per half-verse into the rank tree, segment into phrases. Validate against the common phrase groups (§1.5.5) as test vectors.
8. Practice-loop design (from §5): phrase-level looping, slow playback, tikkun progression (vowels+trope → two-column → STaM only), lookalike-pair drills, 15–20-minute sessions, per-pasuk mastery tracking.
Stage 2 — Ship with a recorded voice (fast, authentic). 9. Record a reference reader (or licence existing recordings — terms unverified) per aliyah; build the word-level alignment layer (forced alignment, hand-checked per aliyah). 10. MVP: pick a parsha → pick an aliyah → hear it leined in Chabad trop with word highlighting, phrase looping, slow playback, tikkun view. This is the tool the user asked for, and it works with zero ML risk.
Stage 3 — Prototype the synthetic voice (in parallel, no deadline). 11. Fine-tune a small Hebrew TTS (Phonikud/Piper-style) on Ashkenazi chanted speech; drive it with the validated motif table via per-phoneme duration + per-frame pitch overrides (the tiny-voca mechanism). 12. A/B the synthetic voice against the recorded reference with a cantor's ear. Ship it only when it is indistinguishable-or-better on melody accuracy — a slightly-wrong synthetic chant is worse than a recording for a learner. 13. If it works, the synthetic voice adds: arbitrary verse ranges, any speed/key, tiny offline size, and the rare tropes for free.
What not to do: don't use Avery/Binder (Reform) or Wolff/Portnoy charts as the Chabad melody — different voicing systems for the same marks. Don't use Sephardi sources at all. Don't ship a synthetic voice before a cantor validates its melody. Don't NFC-normalize the text. Don't treat the §2.5 chart as "the Chabad melody" until Stage 0 step 3 is done.
8. Caveats and thin spots (say plainly)
- The §2.5 motif table is a working model, not the Chabad melody. It is transcribed from a Conservative shul's chart that labels no tradition. It matches the academic Eastern Ashkenazi analysis at every checkpoint, and Chabad is in that family — but a second verification against Slavin is still owed before calling it Chabad.
- "Chabad trop" as a named, distinct system is attested mainly by TropeTrainer's voicing list and community usage. No book or article found defines how it differs, note by note, from the generic Polish-Lithuanian melody. The honest size of the difference is "small, motif-level variants within one family" — with the shalshelet and Slavin's etnachta as the only concrete differences found.
- TropeTrainer's web voices are unverified (recorded vs synthesized). The old Kinnor mobile app used DECtalk TTS; the new site only says "voice styles." Don't cite TropeTrainer as a synthetic-voice precedent without checking.
- Torah Morah's pricing and tradition are unverified beyond "free to download" / likely standard Ashkenazi.
- Sefaria API rate limits were not verified. Bulk-download for a shipped app.
- Mechon Mamre's bulk-download zip URL was not followed — unverified. And its licence bars redistribution without permission.
- The Chabad.org forum details (shalshelet = 2 pazers, karnei parah = mercha kefula + telisha gedola, Slavin's monotone etnachta) are user comments, not verified documentation.
- The canonical written source, Jacobson Chanting the Hebrew Bible (JPS, 2002/2017), was cited throughout but not freely verifiable in this pass. Ch. 6 gives full notation for all six Ashkenazic systems; the companion site is https://www.chantingthehebrewbible.com. Worth consulting before finalising the motif table.
- Exceptions to the trop grammar are numerous (Mechon Mamre). A parser/validator should warn, not reject, on violations.
9. Source list (all links verified 2026-11-17 unless noted)
Taamim system: Wikipedia "Hebrew cantillation" https://en.wikipedia.org/wiki/Hebrew_cantillation · Mechon Mamre / Helmut Richter, "Hebrew cantillation marks and their encoding" https://www.mechon-mamre.org/c/hr/intro.htm · code tables https://www.mechon-mamre.org/c/hr/tables.htm · J101 "Ta'amei HaMikra" https://practice.j101.net/taamei-ha-mikra · Yacavone "Cantillation Reference" (after James D. Price, The Syntax of Masoretic Accents) https://www.yacavone.net/accents/ · Wikipedia "Pashta" https://en.wikipedia.org/wiki/Pashta · Wikipedia "Etnachta" https://en.wikipedia.org/wiki/Etnachta · Iancu, "The Hebrew Cantillation System" (2025) https://musicologypapers.edituramediamusica.ro/images/issues/MP_2025-1_XL-1_04_Alexandra_Rebeca_Iancu.pdf · Malin, "Eastern Ashkenazi Biblical Cantillation" (Yuval 10, 2016) https://jewish-music.huji.ac.il/en/yuval/22542
Chabad melody: TropeTrainer https://www.tropetrainer.com/ · Schmidt "Torah Trope" chart https://www.agudatachim.com/wp-content/uploads/2026/05/torahsheet.pdf · Chabad.org Slavin recordings https://www.chabad.org/multimedia/audio_cdo/aid/983326/jewish/Torah-Reading-Recordings.htm · Chabad.org Interactive Torah Trainer https://www.chabad.org/library/howto/trainer_cdo/aid/2747444/jewish/Interactive-Torah-Trainer-Demo.htm · Chabad.org Torah Reading Trop + forum https://www.chabad.org/multimedia/music_cdo/aid/931078/jewish/Torah-Reading-Trop.htm · Beverly Hills Chabad Schusterman recordings https://www.beverlyhillschabad.com/torah-reading/TORAH-MENU.HTM · ChabadInfo Slavin MP3s http://chabadinfo.com/news/prepare-for-krias-hatorah-parshas-naso/ · Lipinski Bar Mitzvah Torah Trainer (SoundCloud) https://soundcloud.com/themitzvahclub/sets/barmitzvah-torah-trainer-torah-cantillations-taamei-hamikra · Chabad.org "Who Made Up the Way We Sing the Torah?" https://link.chabad.org/library/article_cdo/aid/817346/jewish/Who-Made-Up-the-Way-We-Sing-the-Torah.htm · Jacobson companion site https://www.chantingthehebrewbible.com
Text data: Sefaria developers https://developers.sefaria.org/reference/getting-started · Sefaria v3 texts API https://developers.sefaria.org/reference-link/get-v3-texts · Sefaria-Export https://github.com/Sefaria/Sefaria-Export · tanach.us https://www.tanach.us/ · OSHB/morphhb https://github.com/openscriptures/morphhb · BHSA https://github.com/ETCBC/bhsa · BHSA cantillation docs https://etcbc.github.io/bhsa/cantillation/ · STEPBible TAHOT https://github.com/STEPBible/STEPBible-Data · Mechon Mamre https://www.mechon-mamre.org/c/ct/c0.htm · Hebcal leyning https://github.com/hebcal/hebcal-leyning · Unicode Hebrew table http://www.i18nguy.com/unicode/hebrew.html · hebrew-text-in-practice https://github.com/ohavben/hebrew-text-in-practice · Vayavinu Bamikra https://en.wikisource.org/wiki/Vayavinu_Bamikra
Existing tools: Chabad.org Torah Trainer https://www.chabad.org/library/howto/trainer_cdo/aid/2747444/jewish/Interactive-Torah-Trainer-Demo.htm · TropeTrainer https://www.tropetrainer.com/ · Pocket Torah https://www.pockettorah.com/ (app) · Trope on the Go https://tropeonthego.com/splash · LearnTrope https://www.learntrope.com/ · TorahTrope https://torahtrope.com/pdf/how-to-practice-torah-trope.pdf · IvritSuite Trope Tutor https://ivritsuite.com/trope_tutor.html · Torah Morah https://torahmorah.org/ · Pink Trope https://www.pinkfame.art/ · Sefaria https://www.sefaria.org/texts · Navigating the Bible http://bible.ort.org/
Voice technology: tiny-voca https://huggingface.co/n4ze3m/tiny-voca · tiny-voca code https://github.com/n4ze3m/tiny-voca · tiny-voca demo https://tiny-voca.n4ze3m.com · tiny-voca voice tuner (pitch/duration control) https://muninn.austegard.com/blog/tiny-voca-voice-tuner.html · Phonikud https://phonikud.github.io (paper https://arxiv.org/abs/2506.12311) · BlueTTS https://github.com/maxmelichov/BlueTTS · HebTTS https://github.com/slp-rl/HebTTS (paper https://arxiv.org/pdf/2407.12206) · Samuel / Pink Trombone https://vvolhejn.com/blog/2026-08-17-samuel · YingMusic-Singer https://www.isca-archive.org/interspeech_2026/hao26_interspeech.pdf
Pedagogy: TorahTrope practice guide https://torahtrope.com/pdf/how-to-practice-torah-trope.pdf · Learn to Leyn (Kamionkowski) https://kamionkowski-bet-midrash.teachable.com/p/learn-to-leyn · Torah Reading Guide https://www.torahreadingguide.com/beginners
Full Chabad-melody deep-dive (the §2.5 transcription's method, cross-checks, and full source notes): ~/workspace/research/chabad-trop-research.md.