ComfyUI Node
HeartTranscriptor (Lyrics ASR) 🎤
A ComfyUI node in HeartMuLa with 13 inputs and 1 output.
HeartTranscriptor (Lyrics ASR) 🎤
- pipeline
- audio
- lyrics
◄max_new_tokens256►
◄num_beams2►
◄languageauto►
◄enable_chunkingtrue►
◄tasktranscribe►
◄temperature0.00►
◄no_speech_threshold0.40►
◄compression_ratio_threshold1.8►
◄logprob_threshold-1.0►
◄seed0►
◄condition_on_prev_tokensfalse►
CategoryHeartMuLa
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| pipeline | HEART_TRANSCRIPTOR_PIPELINE | — | |
| audio | AUDIO | — | |
| max_new_tokens | INT | 25616–1024 | Maximum number of new tokens to generate per chunk. |
| num_beams | INT | 21–10 | Number of beams for beam search decoding. |
| language | COMBO | auto | Force generation language. |
| enable_chunking | BOOLEAN | true | Enable chunked processing for long audio. Disable if you encounter errors (e.g. on Mac MPS). |
| task | COMBO | transcribe | 'transcribe' keeps the original language, 'translate' translates to English. |
| temperature | FLOAT | 0.000–1 | Sampling temperature. A fallback sequence (0.0 -> temperature) is built internally. |
| no_speech_threshold | FLOAT | 0.400–1 | Threshold for detecting no-speech segments. |
| compression_ratio_threshold | FLOAT | 1.81–5 | Threshold for compression ratio filtering. |
| logprob_threshold | FLOAT | -1.0-5–0 | Log-probability threshold for filtering low-confidence outputs. |
| seed | INT | 00–18446744073709550000 | Random seed for reproducible results. |
| condition_on_prev_tokensopt | BOOLEAN | false | Whether to condition on previously generated tokens. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| lyrics | STRING | — |