Komisar
55353654b7
- Create modules/omnivoice/ with VoiceAPI, VoiceProfiles, CLI - Add config manager integration with local model support - Add app/komAI.py entry point - Add tests/test_omnivoice.py - Clone OmniVoice to external/ for development - Add omnivoice config to global.yaml
2.6 KiB
2.6 KiB
OmniVoice Module
TTS (Text-to-Speech) модуль на базе OmniVoice.
VENV
Все зависимости в venv:
venv\Scripts\python.exe -m pip install torch==2.8.0+cpu torchaudio==2.8.0+cpu --index-url https://download.pytorch.org/whl/cpu
venv\Scripts\python.exe -m pip install soundfile PyYAML transformers accelerate pydub gradio tensorboardX webdataset librosa
venv\Scripts\python.exe -m pip install -e external/OmniVoice
Использование
Python API
# В venv
venv\Scripts\python.exe
# Инициализация модуля
from modules.omnivoice import register_config, register_logging
register_config()
register_logging()
from modules.omnivoice import get_api, get_profiles
api = get_api()
profiles = get_profiles()
# Voice Cloning
audio = api.clone(
text="Hello, this is a test.",
ref_audio="ref.wav",
ref_text="Reference transcription."
)
# Voice Design
audio = api.design(
text="Hello, this is a test.",
instruct="female, british accent"
)
# Auto Voice
audio = api.auto(text="Hello, this is a test.")
# Сохранение
path = api.save_audio(audio[0], "output.wav")
# Профили
profiles.save_from_generated("my_voice", "Hello", ref_audio="ref.wav")
audio = profiles.generate("my_voice", "Generated text")
CLI
# Скачать модель локально
venv\Scripts\python.exe -m modules.omnivoice.cli download
# Voice Cloning (без --output = воспроизвести)
venv\Scripts\python.exe -m modules.omnivoice.cli clone --text "Hello" --ref-audio ref.wav
venv\Scripts\python.exe -m modules.omnivoice.cli clone --text "Hello" --ref-audio ref.wav --output hello.wav --profile my_voice
# Voice Design
venv\Scripts\python.exe -m modules.omnivoice.cli design --text "Hello" --instruct "female, british accent"
# Auto Voice
venv\Scripts\python.exe -m modules.omnivoice.cli auto --text "Hello"
# Профили
venv\Scripts\python.exe -m modules.omnivoice.cli profiles
venv\Scripts\python.exe -m modules.omnivoice.cli profile-add --name my_voice --ref-audio ref.wav
venv\Scripts\python.exe -m modules.omnivoice.cli profile-remove --name my_voice
venv\Scripts\python.exe -m modules.omnivoice.cli profile-use --profile my_voice --text "Hello"
Конфигурация
Параметры в config/global.yaml:
# omnivoice
omnivoice.model_name: "k2-fsa/OmniVoice"
omnivoice.device: "cuda:0"
omnivoice.dtype: "float16"
omnivoice.num_steps: 32
omnivoice.speed: 1.0
omnivoice.profiles_dir: "data/voice_profiles"
omnivoice.output_dir: "output/voice"
omnivoice.enabled: true