MiniMax-H3-Voice-Preprocessor-Runtime / voice-preprocessing-lock.json
GingerLabsPlatform's picture
Add files using upload-large-folder tool
5b65000 verified
Raw History Blame Contribute Delete
2.53 kB
{
"asr": {
"forced_aligner": false,
"repository": "Qwen/Qwen3-ASR-0.6B",
"revision": "5eb144179a02acc5e5ba31e748d22b0cf3e303b0",
"weight_bytes": 1876091704,
"weight_file": "model.safetensors",
"weight_sha256": "79d6cbd4c98c7bbffe9db2edac07f56cd6637d0d5944b27f6c2b8353840323ea"
},
"contract_version": "h3-voice-preprocessing-v1",
"fish": {
"large_files": [
{
"bytes": 1871099728,
"path": "codec.pth",
"sha256": "74fc41c5a7151c6f350af8bd7e5d6e3accfcc7f3dfbfac23afd35af07052bb2f"
},
{
"bytes": 4986872984,
"path": "model-00001-of-00002.safetensors",
"sha256": "c4218e8ac93be83b35eee30b4f94cb2e9b5ecff40f3e21611438d2f4f8804aad"
},
{
"bytes": 4136876104,
"path": "model-00002-of-00002.safetensors",
"sha256": "76738d23465deaac431433232c0762908cc99a6eddc3d49f67307d92680827be"
}
],
"model_option": "s2-pro-bnb-nf4",
"model_repository": "fishaudio/s2-pro",
"model_revision": "1de9996b6be38b745688de084d87a5633f714e4e",
"node_repository": "https://github.com/Saganaki22/ComfyUI-FishAudioS2.git",
"node_revision": "521f33fe79c081da314dc905ce399c62edb24749",
"quantization": "bitsandbytes NF4 applied at load time",
"workflow": "workflows/fish_s2_reference.api.json",
"workflow_sha256": "9af5b686f4a3d0c8d29759ab7783cf4cba0aa726387596e2213bf771ed62c2dc"
},
"normalization": {
"buckets": [
{
"h3_target_frames": 73,
"name": "short",
"seconds": 3.072,
"waveform_frames": 98304
},
{
"h3_target_frames": 124,
"name": "long",
"seconds": 5.184,
"waveform_frames": 165888
}
],
"format": "stereo PCM16 WAV",
"regenerate_oversized_clips": false,
"sample_rate": 32000,
"selection": "Use the shortest bucket that contains the clip after silence trimming. Pad shorter clips and truncate clips longer than the long bucket.",
"trim_leading_and_trailing_silence": true
},
"prompt_set": {
"clip_count": 10,
"default_language": "en",
"path": "prompts/audio_dataset.v1.json",
"sha256": "e612f5ab9922198bd6634402e51198272d8b15f54b3b7061b7dbc1961fc586f5",
"supported_languages": [
"en",
"ko",
"ja",
"zh"
],
"version": "audio-dataset-v1"
},
"seed_policy": "Use the first 31 bits of SHA-256(train_id) as a base and add the zero-based script index modulo 2^31, yielding ten deterministic distinct seeds."
}