Unified speech inference, training, and optimization

VoiceHub provides one API for text-to-speech (TTS), speech recognition (ASR), and voice activity detection (VAD). It supports Python 3.10–3.12.

Install

Clone the source repository and install the library:

git clone https://github.com/kadirnar/voicehub.git
cd voicehub
python -m pip install .

Install the correct PyTorch build for your hardware first.

Models

Every registered model has a dedicated page in the model list.

from voicehub import AutoModelForTextToSpeech, TTSGenerationConfig

model = AutoModelForTextToSpeech.from_pretrained(
    "parler-tts/parler-tts-mini-v1",
    model_type="parlertts",
    device="cuda",
)
output = model.generate(
    "VoiceHub uses one predictable speech model API.",
    generation_config=TTSGenerationConfig(output_file="speech.wav", seed=42),
)
print(output.file_path)

See the TTS capabilities and ASR/VAD support tables for task-specific inputs.

Train

from voicehub import get_training_spec

spec = get_training_spec("dia")
print(spec.support.value, spec.family_name)

Use the training guide and training support matrix.

Optimize

from voicehub import TTSOptimizationConfig

result = model.optimize(
    TTSOptimizationConfig(
        attn_implementation="auto",
        kernel_backend="auto",
        compile="auto",
    )
)
print(result.manifest())

Use the optimization catalog and TTS benchmark evidence on the target hardware.

Documentation

Read the installation guide, quickstart, and API reference.

VoiceHub is Apache-2.0. Check each checkpoint's separate license before use.