ChatTTS, but with voice gacha and zero-install Colab runs

What it does ChatTTS_colab is essentially a deployment and interface layer over the base ChatTTS text-to-speech model. It bundles the model into a Gradio web UI that runs inside a Colab notebook, so you can generate speech in a browser without wrestling with local CUDA dependencies. The wrapper also adds utilities for long-form audio, basic text normalization for numbers and punctuation, and a multi-role dubbing mode that can split dialogue by character.

The interesting bit The standout feature is “voice gacha” (音色抽卡): it batch-generates random speaker embeddings so you can audition voices lottery-style and save the ones you like. There is even a companion speaker library project that curates stable voices by gender and age, which is a level of voice curation you do not often see in open-source TTS wrappers.

Key highlights

• One-click Colab execution with automatic Gradio public URL tunneling; a Windows offline bundle exists for air-gapped runs
• “Voice gacha” mode for batch-generating and auditioning random voices, plus a separate stable speaker database filterable by gender/age
• Streaming playback that starts playing audio chunks while inference is still running
• Multi-role dubbing mode that assigns different voices to characters and can generate scripts via LLM integration
• Local model loading via --source local for users who cannot reliably download weights from Hugging Face

Caveats

• macOS installation is manual and requires conda/git setup, unlike the advertised one-click Colab flow
• First-run model downloads from Hugging Face can fail silently; the README suggests manually clearing the Hugging Face cache and retrying
• It is a wrapper, not a new model: all synthesis quality and licensing constraints are inherited from the upstream ChatTTS project

Verdict Content creators and Chinese-language developers who want ChatTTS without the environment headaches should try this. Pass if you need a fully self-contained TTS engine with commercial-grade documentation.