ComfyUI Node
LSDub Align
Time every sung word: speech-to-text, reconciled with your lyrics, then forced alignment; builds the song's viseme track with vocalisations (ooh, da da) marked to skip.
LSDub Align
- project
- song
- lyrics
- vocal
- align
- lyric_check
◄pronunciations►
◄stt_modellarge-v3►
◄transcript_json►
CategoryLSDub/audio
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| project | LSDUB_PROJECT | The LSDub project (its fps sets the viseme track rate). | |
| song | LSDUB_SONG | The song. Its lead vocal is used when present, else the vocal, else the mix. | |
| pronunciations | STRING | Words the dictionary lacks, one per line: word=PH ON ES (ARPAbet), e.g. zorbly=Z AO R B L IY. | |
| stt_model | COMBO | large-v3 | Speech-to-text model size. Downloaded on first use. large-v3 is the most accurate; smaller is faster. |
| lyricsopt | LSDUB_LYRICS | The lyrics as sung. Optional: without them the words speech-to-text hears are used (less accurate on sung vocals). | |
| vocalopt | AUDIO | The lead vocal (from LSDub Lead Vocal). Best results: a dry lead vocal. | |
| transcript_jsonopt | STRING | Optional: a saved speech-to-text JSON for this vocal (skips that step). |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| align | LSDUB_ALIGN | Word timings + the song viseme track. |
| lyric_check | STRING | Markdown: least certain words and the skipped vocalisations. |