110b2298b7
- stt_local: Whisper large-v3-turbo MLX, transcript + timestamped json - lipsync_local: Rhubarb visemes (CPU lane), dialog-guided - voice_clone_local: Chatterbox on MPS (needs setuptools<81 for perth/pkg_resources) - wan_video: Wan 2.2 TI2V-5B t2v/i2v via resident ComfyUI, frames->mp4 - llm_local: Qwen3-30B-A3B-4bit dialogue writer (primary-only) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
18 lines
662 B
JSON
18 lines
662 B
JSON
{
|
|
"id": "stt_local",
|
|
"name": "Transcribe (local, Whisper)",
|
|
"category": "audio",
|
|
"description": "Audio → transcript (.txt + timestamped .json) entirely on this Mac via Whisper large-v3-turbo on MLX. Subtitles, VO QA round-trips (does the generated line transcribe back correctly?), dictation. Needs ffmpeg on PATH for non-wav inputs.",
|
|
"accepts": ["audio"],
|
|
"produces": ["text"],
|
|
"resources": "gpu",
|
|
"entry": "run.py",
|
|
"python": "venvs/whisper/bin/python",
|
|
"params_schema": {
|
|
"type": "object",
|
|
"properties": {
|
|
"language": {"type": "string", "default": "en", "description": "ISO language code; empty = autodetect"}
|
|
}
|
|
}
|
|
}
|