Breeze TTS 2

Turn text into natural speech with Breeze TTS 2, an open source text-to-speech model with zero-shot voice cloning. Provide a short reference clip, and the model speaks your script in that voice—no retraining required. Released for research and development by the Baidu breeze-tts team.

Try Breeze TTS 2 Demo View on GitHub

Approach: Zero-Shot TTS
Cloning Input: Short Reference
Output: Speech Audio
License: Open Source

Experience Breeze TTS 2 Live Demo

Type a script, attach a short reference clip, and hear the model speak it back in the cloned voice. Try different scripts, speakers, and lengths to hear how Breeze TTS 2 handles pacing, tone, and natural pauses.

Core Capabilities

Natural speech from a simple workflow

Breeze TTS 2 focuses on one task and does it directly: turn your text and a short voice reference into natural spoken audio you can review and use.

Zero-Shot Voice Cloning: Breeze TTS 2 clones a voice from a short reference clip without any model retraining. The model captures the timbre and speaking style of the reference and applies it to entirely new text. This makes it practical for quick drafts, prototypes, and personalizing narration without recording a full voice dataset.
Natural Prosody and Pacing: The model generates speech with natural pacing, pauses, and emphasis rather than flat, robotic delivery. Long sentences are handled with appropriate phrasing, so narration reads smoothly for scripts, articles, and dialogue-style content. As with any TTS model, reviewing the output before publishing is recommended.
Text-Driven Workflow: Everything starts from your script. Paste or type the text you want spoken, attach a reference clip, and generate audio. This straightforward input-to-output workflow fits narration, video voiceovers, e-learning drafts, and accessibility use cases where you already have the words ready.
Open Source Code and Weights: Breeze TTS 2 is released as an open source project, with the code available on GitHub and model weights published for download. You can inspect the pipeline, run it on your own hardware, and integrate it into your own applications instead of depending on a hosted API.
Self-Hosted Deployment: Because the model can run locally, you keep control over where your scripts and reference audio are processed. This matters for teams working with unpublished material, client voice samples, or privacy-sensitive content. Check the project repository for current hardware requirements before deploying.
Iterate by Re-Generating: If a take does not sound right—wrong pacing, odd emphasis, or an unnatural pause—you can adjust the text or reference and generate again. Each run is non-destructive: the script and reference stay on file, so iterating toward the read you want is fast and repeatable.

Common Questions

What is Breeze TTS 2?: Breeze TTS 2 is an open source text-to-speech model from the Baidu breeze-tts team. It converts written text into natural speech and can clone a voice from a short reference clip in a zero-shot way, meaning no retraining or fine-tuning is required for a new speaker. The code is available on GitHub with published model weights.
How does zero-shot voice cloning work?: You provide a short audio clip of the voice you want to use, along with the text you want spoken. Breeze TTS 2 extracts the voice characteristics from the reference and generates new speech in that voice for your script. Results are typically better with a clean reference clip that has minimal background noise and a single speaker.
Can I run Breeze TTS 2 on my own hardware?: Yes. Breeze TTS 2 is open source and can be self-hosted. Running the model locally keeps your scripts and reference audio under your own control. Check the GitHub repository for the current hardware requirements and installation steps, as requirements depend on the released model size and inference settings.
Is Breeze TTS 2 free to use?: The model and code are released openly for research and development purposes. Review the license in the official repository before using Breeze TTS 2 in commercial products, since terms such as attribution or usage restrictions may apply. The online demo on this page is provided for quick testing.
What affects the quality of cloned speech?: The clearest factor is reference audio quality: a clean, single-speaker clip without background noise or heavy reverb produces better clones than a noisy or mixed recording. The script matters too—natural sentence structure tends to read more smoothly than dense, unformatted text. If a result sounds off, adjust the reference or script and generate again.
Can I use the generated audio commercially?: It depends on the license of the model and the rights to the voice you clone. Review the license terms in the breeze-tts repository, and only clone voices you have permission to use—typically your own voice or one you have written consent for. If you plan to publish or monetize the output, verify the terms first.