ComfyUI Node
Local Whisper Transcribe
Transcribes ComfyUI AUDIO locally with faster-whisper and releases the model before MiniMax H3 loads.
Local Whisper Transcribe
- audio
- transcript
- segments_json
- detected_language
- status
◄modellarge-v3►
◄languageauto►
◄deviceauto►
◄compute_typeauto►
◄beam_size5►
◄vad_filtertrue►
◄initial_prompt►
CategoryVision Prompt Assistant/Audio
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| audio | AUDIO | — | |
| model | COMBO | large-v3 | 2 options: large-v3, large-v3-turbo |
| language | COMBO | auto | 15 options: auto, en, es, ja, zh, hi, +9 |
| device | COMBO | auto | 3 options: auto, cuda, cpu |
| compute_type | COMBO | auto | auto uses float16 on CUDA and int8 on CPU. int8_float16 uses less VRAM with a small possible accuracy tradeoff. |
| beam_size | INT | 51–10 | — |
| vad_filter | BOOLEAN | true | — |
| initial_promptopt | STRING | Optional vocabulary or context hint; it is not added to the result. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| transcript | STRING | — |
| segments_json | STRING | — |
| detected_language | STRING | — |
| status | STRING | — |