ozguradmin/ahenk-turkish-tts is a collection of tools that turn Turkish text into speech. It holds Ahenk, a large model for GPU, Ahenk Cep, an 8.6-million-parameter model that runs on CPU, trnorm, which normalizes numbers, dates, times, and amounts to Turkish pronunciation, and the TR-Zor test set. It also includes Ahenk Okur, which reads documents and photos aloud in the browser. No separate recent release is registered.
Ahenk Cep is a small speech synthesis model that runs on-device. According to the repository description, it is a student model that learns the voice of the larger model, and it uses an EMA Lightning architecture. Its size is 34MB, and the published real-time factor (RTF) is about 0.2 on two CPU threads. RTF compares the time it takes to synthesize speech against the playback duration. Ahenk and Ahenk Cep speak Turkish with one designed synthetic voice, not the voice of any particular person. The repository is written in Python and uses the Apache-2.0 license.
That said, the published RTF figure alone makes it hard to judge performance differences against other models or actual usage conditions. FACTS does not present measurement results for TR-Zor, and no recent release is registered either. So this repository is useful as a reference for exploring tools related to Turkish synthesis, but the evidence available for checking release-by-release changes or comparative performance is limited.