AI Briefingema-lightning: Turkish TTS Accuracy and Speed Rank First with 8.6M Parameters
canberkkkkkk/ema-lightning
About the project
In the field of Turkish speech synthesis, it outperforms ElevenLabs v4 and Gemini 3.8 with a 0.92% WER. It consists of an ultra-lightweight DiT model and vocoder with a total of 8.6M parameters, and the model file is only 34MB.
It generates audio at 440 times real-time speed on an RTX 4090, with a time-to-first-audio of 3.86ms. In streaming mode, listening is possible simultaneously with text generation, and it operates at 6 times real-time speed even in CPU environments.
The built-in Playhead scheduler batches multiple requests on the GPU, supporting 64 concurrent streams. It can run completely offline, eliminating data leakage concerns, and is freely available for commercial use under the Apache 2.0 license.
canberkkkkkk/ema-lightning
The original page has no description.
text-to-speech
This introduction 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 errors, attribution issues, or removal requests via Contact.