v0.6.0

Expensify/Appv0.6.0Mar 28, 2026by KoljaB

AI Summary

Major feature release adding seven new TTS engines (Faster Qwen3, Cartesia, MiniMax Cloud, ModelsLab, CAMB AI MARS, NeuTTS, PocketTTS), WebSocket streaming support, and multiple engine usability improvements.

Key Highlights

New TTS Engines: Faster Qwen3, Cartesia, MiniMax Cloud, ModelsLab, CAMB AI MARS, NeuTTS, PocketTTS
WebSocket streaming support with multi-user and bidirectional audio
Engine usability improvements (API key parameter, language parameter, volume control)
NeuTTS supports on-device TTS with voice cloning
PocketTTS optimized for CPU with low latency

New Features

Seven new TTS engines with various language and voice support
WebSocket streaming endpoint with Python demo client
MPV audio device selection option
Adjustable volume parameter (0.0-1.0)
Audio format detection and conversion for non-Kokoro engines

Full Release Notes

# RealtimeTTS v0.6.0

### Features

- **New TTS Engines**
  - **Faster Qwen3 TTS**: Added `FasterQwenEngine` engine. See tests/faster_qwen_emotions.py and tests/faster_qwen_test.py for details how to implement. Demo video: https://www.youtube.com/watch?v=ZOKcUpJlrXQ
  - **Cartesia**: Added engine for Cartesia speech synthesis via WebSocket API. ([#348])
  - **MiniMax Cloud TTS**: Added `MiniMaxEngine` for MiniMax T2A v2 API with 2 models (`speech-2.8-hd`, `speech-2.8-turbo`), 12 voice presets, and runtime parameter control. ([#369])
  - **ModelsLab TTS**: Added `ModelsLabEngine` supporting 30+ voices in 9 languages, speed and emotion control, and lazy loading. ([#368])
  - **CAMB AI MARS TTS**: Added `CambEngine` and `CambVoice` for CAMB AI's MARS models, supporting 140+ languages and streaming output. ([#367])
  - **NeuTTS**: Added `NeuTTSEngine` for on-device TTS with voice cloning (3s reference audio, CPU/CUDA/MPS support). ([#359])
  - **PocketTTS**: Added `PocketTTSEngine` for CPU-optimized TTS (Kyutai Labs, 8 voices, voice cloning, low latency). ([#358])

- **WebSocket Streaming**
  - Real-time TTS streaming via WebSocket endpoint with multi-user support, bidirectional audio, and enhanced web UI. Includes Python demo client. ([#356])

### Improvements

- **Engine Usability**
  - Allow `OpenAIEngine` to accept API key as parameter (fallback to env var). ([#361])
  - Add language parameter to `ZipvoiceEngine` for output and prompt speech. ([#362])
  - Add `mpv_audio_device` option to `AudioConfiguration`/`TextToAudioStream` for MPV playback device selection. ([#327])
  - Add adjustable volume parameter (0.0–1.0) to `TextToAudioStream`. ([#335])
  - PiperEngine: Streamline synthesis and add samplerate detection from model config for better quality with larger models. ([#346], [#347])
  - Conditional logging: Only print "SYNTHESIS FINISHED" if logging is enabled. ([#332])

- **General**
  - Install `portaudio` for MacOS. ([#328])

### Fixes

- Correct typo in `requirements.txt` for `pypinyin` version. ([#355])
- Fix missing comma in `__all__` that affected engine exports. ([#367])
- Add audio format detection/conversion for non-Kokoro TTS engines. ([#356])
- Fix voice retrieval errors and improve engine initialization logic. ([#356])

### Other

- Updated documentation for new engines, playback options, and WebSocket usage.
- Added/updated test files and demo scripts for new engines.
- No breaking changes; all updates are backward compatible.

---

**PRs:**
[#369], [#368], [#367], [#362], [#361], [#359], [#358], [#356], [#355], [#348], [#347], [#346], [#335], [#332], [#328], [#327]