Files

191 lines
4.5 KiB
JSON

{
"schema_version": 1,
"family": "kroko_asr",
"display_name": "Kroko Community ASR",
"description": "Native Zipformer2 RNN-T transcription for the public free Kroko Community single-language packages, with offline and stateful streaming inference, greedy and modified beam decoding, hotwords, endpoint segments, partial results, and word timestamps.",
"category": "asr",
"status": "community",
"tasks": [
"asr"
],
"modes": [
"offline",
"streaming"
],
"languages": [
"de",
"en",
"es",
"fr",
"it",
"he",
"nl",
"pt",
"sv",
"tr"
],
"capabilities": {
"asr": [
"word_timestamps",
"partial_results",
"segments"
]
},
"options": {
"request": [
{
"name": "language",
"type": "string",
"description": "Language code matching the selected single-language Kroko package; default auto uses the package language.",
"required": false,
"default": "auto"
},
{
"name": "decoding_method",
"type": "enum",
"description": "RNN-T decoding method; greedy search remains the parity-tested default.",
"values": [
"greedy_search",
"modified_beam_search"
],
"required": false,
"default": "greedy_search"
},
{
"name": "num_beams",
"type": "int",
"description": "Maximum active hypotheses retained by modified beam search.",
"required": false,
"min": 1,
"max": 64,
"default": 4
},
{
"name": "blank_penalty",
"type": "float",
"description": "Non-negative score subtracted from the RNN-T blank logit.",
"required": false,
"min": 0.0,
"default": 0.0
},
{
"name": "hotwords",
"type": "string",
"description": "Slash- or newline-separated natural-text phrases; requires modified beam search.",
"required": false
},
{
"name": "hotwords_score",
"type": "float",
"description": "Non-negative per-token context boost for hotword phrases.",
"required": false,
"min": 0.0,
"default": 1.5
},
{
"name": "enable_endpoint",
"type": "bool",
"description": "Enable automatic endpoint speech segments.",
"required": false,
"default": false
},
{
"name": "rule1_min_trailing_silence_sec",
"type": "float",
"description": "Endpoint timeout in seconds even when no speech token was decoded.",
"required": false,
"min": 0.0,
"default": 2.4
},
{
"name": "rule2_min_trailing_silence_sec",
"type": "float",
"description": "Endpoint silence in seconds after a speech token was decoded.",
"required": false,
"min": 0.0,
"default": 1.2
},
{
"name": "rule3_min_utterance_length_sec",
"type": "float",
"description": "Maximum utterance duration in seconds before an endpoint.",
"required": false,
"min": 0.0,
"default": 20.0
}
],
"session": [],
"load": []
},
"runtime": {
"tags": [
"gguf",
"stream"
]
},
"sources": [
{
"format": "gguf",
"roots": {
"model": ".",
"weights": "$gguf"
},
"files": {
"config": "model:config.json",
"tokens": "model:tokens.txt"
},
"tensors": {
"weights": "weights:"
}
},
{
"format": "safetensors",
"roots": {
"model": "."
},
"files": {
"config": "model:config.json",
"tokens": "model:tokens.txt"
},
"tensors": {
"weights": "model:model.safetensors"
}
}
],
"package_defaults": {
"download": {
"kind": "huggingface_snapshot",
"repo": "audio-cpp/audio.cpp-gguf",
"revision": "main",
"gated": false
}
},
"packages": [
{
"id": "kroko_asr_community_q8_0",
"display_name": "Kroko Community ASR Q8_0 GGUF",
"default": true,
"format": "gguf",
"precision": "q8_0",
"target_directory": "Kroko-ASR-GGUF",
"files": [
"Kroko-ASR-GGUF/kroko-en-community-64-l-q8_0.gguf"
],
"strip_prefix": "Kroko-ASR-GGUF"
}
],
"dependencies": [],
"ui": {
"recommended_package": "kroko_asr_community_q8_0",
"tags": [
"ASR",
"GGUF"
],
"docs": [
"docs/asr.md",
"docs/community_models/kroko_asr.md",
"docs/gguf.md"
]
}
}