Try it now

Fast, private text to speech.

Natural voices in six languages, generated right in your browser.

Checking for the model…

In browser

Hear it switch languages mid-sentence.

Two voices, five languages, one conversation. Recorded with an earlier version of the model.

Turn any text into a voice note.

Message BlueTTS from the chat app you already use. No signup, nothing to install.

0,000 messages read aloud for 0,000 people so far

MIT licensed, top to bottom.

The models, the voices and this browser runtime are all on GitHub. A few lines of Python get you started.

Python, from the README
import soundfile as sf
from blue_onnx import BlueTTS

tts = BlueTTS(onnx_dir="onnx_models", style_json="voices/noa.json")
samples, sr = tts.synthesize("שלום, זהו מודל דיבור בעברית.", lang="he")
sf.write("out.wav", samples, sr)

Speed

Times faster than real time

  • NVIDIA RTX 30901500×
  • Apple M470×
  • AMD Ryzen 9 7950X35×

In the repo

  • ONNX models
  • Four voices: Noa, Lily, Daniel and Adam
  • Zero-shot voice cloning, no fine-tuning
  • The in-browser runtime: WebGPU, with a CPU fallback
View on GitHub Model weights on Hugging Face

MamboTTS, BlueTTS on your own machine

A separate desktop app for Hebrew and English speech, on macOS, Windows and Linux.

MamboTTS: Hebrew text in the editor, the Rotem and Roi voices in the sidebar, and a generated clip ready to play.
  • Fully offline after the model downloads
  • Runs on an ordinary CPU, no GPU required
  • Rotem and Roi voices, instant preview, WAV export
  • Hebrew diacritics and IPA editor
  • CLI, local HTTP API with Swagger docs, and /skill instructions for agents
Download MamboTTS MamboTTS on GitHub

Questions, bugs or ideas?

Write to me. I read everything.

Email me LinkedIn