MusicXML: expose lyric <syllabic> (begin/middle/end) on imported beats

[weblings]

Note

AI-authored disclosure (alphatab-ai-authored-v1)

Portions of this content were authored by an AI agent. The agent has read
AGENTS.md and the human submitter accepts responsibility for
compliance with the rules in that document.

I'm importing MusicXML with alphaTab and then reading lyrics from the resulting Score. MusicXML marks whether a syllable continues a word with <syllabic> (single, begin, middle, end). After import, that information isn't available: each beat has its lyric text, but there's no way to tell that "Tra" / "la" / "la," belong to one word rather than being separate words. The test file 61b-MultipleLyrics.xml has this on every verse (verse 1 is begin / middle / end / single / begin / end).

What I'd like: some way to read each lyric's syllabic value (per verse) from the imported model, so a consumer can rebuild words from syllables.

Why: I convert imported scores into another format that needs word boundaries (line-wrapping on whole words). Without it, multi-syllable words can't be distinguished from consecutive one-syllable words.

I have a working approach locally but didn't want to propose an API shape unasked. Since this touches the public model, I'd like to hear what shape you'd prefer (and whether it's wanted at all) before offering anything.

Checked against develop c53d5ad (alphaTab 1.9.0): lyrics come through as plain per-verse strings, e.g. ["Tra","Tra","TRA"] for 61b.

[Danielku15]

An interesting usecase. can you explain further how you would use the infromation exactly. at this stage your proposal is quite low level. It would help me to understand how you later on plan to use the information to design the feature properly. e.g. what format do you export again and what's the role of alphaTab in your software.

We have #2908 on the table for 1.10 and we could make some additional information available there.

Parsing and storing this information comes at a cost (especially memory) so I want to be sure we're not generating unnecessary overhead for the standard usecase. In best case alphaTab would also later make use of this information to render lyrics accordingly.

[weblings]

Note

AI-authored disclosure (alphatab-ai-authored-v1)

Portions of this content were authored by an AI agent. The agent has read
AGENTS.md and the human submitter accepts responsibility for
compliance with the rules in that document.

Thanks! Some more context:

How we use alphaTab

• Currently import only, no rendering: Guitar Pro and MusicXML go in, and we walk the Score to build a rhythm-game chart.
• Lyrics become timed entries, and lines wrap on whole words, so we need to know which syllables belong to the same word.

The gap: the same lyric gives different beats depending on the format (1.9.0, verse 1 of 61b)

Source	beat.lyrics[0]
.gp file with lyric text Tra-la-la, ja!	"Tra-" "la-" "la," "ja!"
61b-MultipleLyrics.xml (begin/middle/end/single)	"1.Tra" "la" "la," "ja!"

• GP keeps the word join in the text: the Lyrics parser keeps the dash, as Lyrics.test.ts covers ('AAA-BBB' → ['AAA-', 'BBB']).
• MusicXML drops <syllabic>, so one three-syllable word looks the same as three separate words.
• Export carries the gap forward: 61b → Gp7Exporter → reload still gives "1.Tra" "la" "la,". With no dashes on the beats, the GP file has three separate words where the .gp row above has one.

Proposal: no new model data

• For begin/middle syllables, the MusicXML importer adds - to the text, matching what GP gives today.
• No new field and no extra memory, since it's the same string arrays.
• MusicXML and GP lyrics then look the same for rendering, export and integrators.
• For Expose Lyrics in public api #2908: rebuilding full lyric text from MusicXML needs exactly this information. Join with no space after a - syllable and with a space otherwise; the existing parser then gives the same chunks back. So this could ship as part of Expose Lyrics in public api #2908.
• Cost: MusicXML lyrics would render with the dash, as GP lyrics already do, so existing MusicXML lyric renders would change.

Fallback if you'd rather keep the text as written

• A single continuesWord-style flag on the per-syllable object Expose Lyrics in public api #2908 exposes. We only need "continues into the next one"; begin vs. middle doesn't matter to us.
• A per-beat array of syllabic strings (what I tried locally) is the heaviest option and the one alphaTab would use least, so I wouldn't push for it.

Glad to go whichever way fits #2908.

Checked against npm 1.9.0 and develop 7d3efe7.