AI BriefingC++ Port of Echo-TTS Released
·2026.05.07 03:43
Key point
A C++ port of Echo-TTS that supports efficient inference using GGML and ONNX Runtime has been released.
Details
A C++ port of the multi-speaker Echo-TTS, which supports speaker reference conditioning, has been released. This implementation uses GGML to run the Diffusion Transformer and ONNX Runtime to handle the DAC autoencoder, operating in CUDA-based GPU environments.
Key Features:
- Model size: approximately 3.3GB with Q8 quantization, approximately 5.6GB for the F16 version
- Server mode: OpenAI-compatible server mode with chunking support
- Various features: multi-voice support using reference WAV files, Euler sampling, blockwise generation, and continuous mode support
- Convenience: portable ZIP file provided including CUDA 12.8, cuDNN 9.21, and ONNX Runtime
It has currently been tested on Windows, and the code and models are available via GitHub and Hugging Face.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Jevper Released: A Jev Classification Wrapper for OpenAI-Compatible Models with Automatic Logprobs Detection
Rootless Linux Sandbox 'Drop' Released for Isolating AI Agents and Third-Party Programs