Skip to content

Getting started

Welcome to VoiceHub Docs!

A shared Python interface for speech synthesis, transcription, and voice activity detection.

Why VoiceHub?

Work with 68 integrations: 34 TTS backends, 23 ASR providers, and 11 VAD providers, each with a documented configuration, model, and processor contract.

  • One model lifecycle — discover, configure, load, and run speech models through consistent APIs.
  • Architecture-aware behavior — preserve each model's conditioning, codecs, decoding, and outputs.
  • Reproducible comparisons — inspect checkpoint provenance, inference settings, timings, and measured differences.
  • Explicit support — check each integration's requirements and tested limits before choosing a model.

Explore the library architecture →

Get started

  • Explore models

    Browse TTS, ASR, and VAD integrations, checkpoints, and their capabilities.

    Explore models →

  • Inference guides

    Install VoiceHub and run your first speech request with the shared API.

    Start generating →

  • Fine-tuning

    Check training support and prepare data for your selected architecture.

    Fine-tuning guide →

  • Examples

    Follow runnable notebooks for inference, data preparation, and training.

    Open examples →

Check the evidence

A registered model is not proof of upstream quality or performance parity. The upstream comparison guide distinguishes actual paired measurements, failures, and outstanding evaluations. Waveform agreement, transcript error rates, and inference timing are separate checks.

Checkpoint weights are downloaded lazily. Training tools are available through voicehub[training]. VoiceHub's license covers the library; integrated source, checkpoints, codecs, and datasets may have separate terms.