0.4.2

bknd-io/bknd0.4.2May 19, 2025by edwko

AI Summary

Focused on high-throughput audio processing by introducing batched decoding interfaces and fade-in/fade-out effects. Added support for the OuteTTS 1.0 0.6B model and new backends like EXL2 Async and VLLM.

Key Highlights

Fade-in/fade-out audio decoding to eliminate clipping artifacts
Batched decoding interfaces (EXL2 Async, VLLM, llama.cpp)
Support for OuteTTS 1.0 0.6B model
Single-stream decoding for llama.cpp server

New Features

Batched inference support
Audio artifact reduction
New model compatibility
Enhanced decoder backends

Full Release Notes

# OuteTTS v0.4.2

* **Fade-in / Fade-out Audio Decoding**
  Introduced quick fade-in and fade-out on decoded audio chunks to eliminate clipping artifacts at segment boundaries.

* **Batched Decoding Interfaces**
  Added support for high-throughput, batched inference via three new backends:

  * **EXL2 Async**: Asynchronous batch processing using the EXL2.
  * **VLLM**: Asynchronous batch decoding with VLLM. (Experiment support)
  * **llama.cpp Async Server Endpoint**: Connects to a continuously-batched llama.cpp server for async inference.

* **Single-Stream Decoding**

  * **llama.cpp Server Endpoint**: single-stream decode endpoint for llama.cpp server.

* **OuteTTS 1.0 0.6B Model Support**
  Compatibility with the new [OuteTTS-1.0-0.6B](https://huggingface.co/OuteAI/OuteTTS-1.0-0.6B), including config defaults.

* **Batched Interface Parameters**
  New configuration options to control batched interface.

* Enhanced pre-prompt normalization pipeline.

* Documentation Updates, expanded the batched interface usage.