230 lines
6.6 KiB
JSON
230 lines
6.6 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"family": "dramabox",
|
|
"display_name": "DramaBox",
|
|
"description": "DramaBox is an English expressive TTS and voice-cloning model packaged for audio.cpp as a standalone GGUF bundle. It combines Gemma text conditioning, diffusion sampling, reference-audio conditioning, long-form chunking, and 48 kHz stereo output.",
|
|
"category": "tts",
|
|
"status": "experimental",
|
|
"tasks": [
|
|
"tts",
|
|
"clone"
|
|
],
|
|
"modes": [
|
|
"offline"
|
|
],
|
|
"languages": [
|
|
"en"
|
|
],
|
|
"runtime": {
|
|
"tags": [
|
|
"gguf"
|
|
]
|
|
},
|
|
"capabilities": {
|
|
"tts": [
|
|
"speaker_reference",
|
|
"style_control",
|
|
"long_form"
|
|
],
|
|
"clone": [
|
|
"speaker_reference",
|
|
"style_control",
|
|
"long_form"
|
|
]
|
|
},
|
|
"options": {
|
|
"request": [
|
|
{
|
|
"name": "target_voice",
|
|
"type": "audio_path",
|
|
"description": "Reference voice WAV path for voice cloning. If omitted, DramaBox generates from text without reference-audio conditioning.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "negative_prompt",
|
|
"type": "string",
|
|
"description": "Negative text conditioning used when guidance_scale enables classifier-free guidance; omitted uses the built-in quality prompt.",
|
|
"required": false
|
|
},
|
|
{
|
|
"name": "duration_sec",
|
|
"type": "float",
|
|
"description": "Explicit target duration in seconds; default 0 uses the prompt-duration estimator.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 0.0
|
|
},
|
|
{
|
|
"name": "num_inference_steps",
|
|
"type": "int",
|
|
"description": "Diffusion sampling step count; default comes from config.json, 30 in the current package.",
|
|
"required": false,
|
|
"min": 1,
|
|
"default": 30
|
|
},
|
|
{
|
|
"name": "guidance_scale",
|
|
"type": "float",
|
|
"description": "Classifier-free guidance scale; default comes from config.json, 2.5 in the current package. Values greater than 1 enable CFG.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 2.5
|
|
},
|
|
{
|
|
"name": "spatio_temporal_guidance_scale",
|
|
"type": "float",
|
|
"description": "Spatio-temporal guidance scale; default comes from config.json, 1.5 in the current package. Values greater than 0 enable STG.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 1.5
|
|
},
|
|
{
|
|
"name": "duration_scale",
|
|
"type": "float",
|
|
"description": "Multiplier applied to the estimated prompt duration when duration_sec is 0; default comes from config.json, 1.1 in the current package.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 1.1
|
|
},
|
|
{
|
|
"name": "reference_duration_sec",
|
|
"type": "float",
|
|
"description": "Reference voice crop/repeat duration in seconds; default comes from config.json, 10.0 in the current package.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 10.0
|
|
},
|
|
{
|
|
"name": "guidance_rescale",
|
|
"type": "string",
|
|
"description": "Guidance rescale value. The default auto mode derives a rescale value from guidance_scale; a numeric string requests an explicit value.",
|
|
"required": false,
|
|
"default": "auto"
|
|
},
|
|
{
|
|
"name": "audio_chunk_threshold_sec",
|
|
"type": "float",
|
|
"description": "Estimated duration threshold that switches a request to long-form chunking; default 45.0 seconds.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 45.0
|
|
},
|
|
{
|
|
"name": "audio_chunk_duration_sec",
|
|
"type": "float",
|
|
"description": "Target estimated duration for each long-form chunk; default 37.0 seconds.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 37.0
|
|
},
|
|
{
|
|
"name": "cross_fade_duration_sec",
|
|
"type": "float",
|
|
"description": "Equal-power cross-fade between long-form chunks in seconds; default 0.05.",
|
|
"required": false,
|
|
"min": 0.0,
|
|
"default": 0.05
|
|
},
|
|
{
|
|
"name": "seed",
|
|
"type": "int",
|
|
"description": "Torch-compatible CUDA noise seed for diffusion sampling; default 42.",
|
|
"required": false,
|
|
"min": 0,
|
|
"default": 42
|
|
}
|
|
],
|
|
"session": [
|
|
{
|
|
"name": "perf_mode",
|
|
"type": "enum",
|
|
"description": "Attention implementation mode. Default off keeps the exact reference-query attention path; flash_attention enables the optimized path.",
|
|
"values": [
|
|
"off",
|
|
"flash_attention"
|
|
],
|
|
"required": false,
|
|
"default": "off"
|
|
},
|
|
{
|
|
"name": "mem_saver",
|
|
"type": "bool",
|
|
"description": "Release staged runtime graphs and weights immediately after each request phase to reduce peak and resident VRAM; default false keeps components cached for later reuse.",
|
|
"required": false,
|
|
"default": false
|
|
}
|
|
],
|
|
"load": []
|
|
},
|
|
"package_defaults": {
|
|
"download": {
|
|
"kind": "huggingface_snapshot",
|
|
"repo": "audio-cpp/audio.cpp-gguf",
|
|
"revision": "main",
|
|
"gated": false
|
|
}
|
|
},
|
|
"packages": [
|
|
{
|
|
"id": "dramabox_q8_0",
|
|
"display_name": "DramaBox Q8_0 GGUF",
|
|
"default": true,
|
|
"format": "gguf",
|
|
"precision": "q8_0",
|
|
"target_directory": "DramaBox-GGUF",
|
|
"files": [
|
|
"DramaBox-GGUF/dramabox-q8_0.gguf"
|
|
],
|
|
"strip_prefix": "DramaBox-GGUF"
|
|
}
|
|
],
|
|
"dependencies": [],
|
|
"ui": {
|
|
"recommended_package": "dramabox_q8_0",
|
|
"tags": [
|
|
"TTS",
|
|
"Clone",
|
|
"GGUF"
|
|
],
|
|
"docs": [
|
|
"docs/tts.md",
|
|
"docs/gguf.md"
|
|
]
|
|
},
|
|
"sources": [
|
|
{
|
|
"format": "gguf",
|
|
"roots": {
|
|
"model": ".",
|
|
"weights": "$gguf"
|
|
},
|
|
"files": {
|
|
"config": "model:config.json",
|
|
"audio_components_config": "model:audio_components_config.json",
|
|
"gemma_config": "model:gemma-3-12b-it-bnb-4bit/config.json",
|
|
"gemma_tokenizer_model": "model:gemma-3-12b-it-bnb-4bit/tokenizer.model",
|
|
"gemma_tokenizer_json": "model:gemma-3-12b-it-bnb-4bit/tokenizer.json",
|
|
"gemma_tokenizer_config": "model:gemma-3-12b-it-bnb-4bit/tokenizer_config.json"
|
|
},
|
|
"tensors": {
|
|
"dit_weights": {
|
|
"source": "weights:",
|
|
"prefix": "dit"
|
|
},
|
|
"audio_weights": {
|
|
"source": "weights:",
|
|
"prefix": "audio"
|
|
},
|
|
"gemma_weights": {
|
|
"source": "weights:",
|
|
"prefix": "gemma"
|
|
},
|
|
"silence_latent": {
|
|
"source": "weights:",
|
|
"prefix": "silence"
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|