v1.0.4

KoljaB/RealtimeSTTv1.0.4Aug 20, 2026by KoljaB

AI Summary

This release introduces a production-ready FastAPI server with TLS and authentication, adds support for Nemotron 3.5 and Parakeet TDT models via sherpa-onnx, and enhances the streaming engine with word timestamps and custom APIs. Additionally, it resolves critical stability issues related to WebSocket handling, chunk boundaries, and lifecycle management.

Key Highlights

Production-ready FastAPI HTTP/WebSocket server with TLS, authentication, and backpressure.
Added fast CPU streaming with Nemotron 3.5 ASR and Parakeet TDT models.
New `stt-install-sherpa-models` command for verified model installation.
Improved streaming stability with fixes for WebSocket turns, chunk-boundary inconsistencies, and lifecycle leaks.

New Features

FastAPI HTTP/WebSocket server with security features (TLS, auth, backpressure).
Nemotron 3.5 ASR and Parakeet TDT transcription support.
Verifiable model installation via `stt-install-sherpa-models`.
Custom engine APIs, word timestamps, and `feed_audio_file()` functionality.

Full Release Notes

# RealtimeSTT v1.0.4

### Major Features

- Added fast CPU streaming with Nemotron 3.5 ASR and Parakeet TDT final refinement through sherpa-onnx.
- Added `stt-install-sherpa-models` for verified model installation.
- Added a production-ready FastAPI HTTP/WebSocket server with TLS, authentication and backpressure.
- Added new transcription engines, custom engine APIs, word timestamps and `feed_audio_file()`.

### Fixes

- WebSocket turns now produce exactly one authoritative final followed by one completion.
- Fixed chunk-boundary inconsistencies, late-event races, streaming buffer growth and lifecycle leaks.
- Improved realtime partial stability, cancellation and slow-client handling.

[Full release notes](https://github.com/KoljaB/RealtimeSTT/blob/master/RELEASE_NOTES.md)