BHOSAI/SARA_TTS

Voice of SARA

Baku Higher Oil School Research and Development Center on AI introduce their new Text-to-Speech model in collaboration with PRODATA. Model is based on VITS architecture, referenced to Meta MMS on Azerbaijani.

(c) Image has been generated by using Microsoft AI Image Generator!

Meta MMS model has good performance in naturalness of the speech while it was not robust to the change in the input tokens. Intonation varied according to the input tokens.

Our team has built speech and text pairs from public sources and combined them with 2-3 sentences to create continuous speech in the input.

Thanks to Kavsar Huseynova, Elvin Mammadov, Qurban Quliyev and PRODATA for the the contributions to this project!

All rights are reserved!

Note: Team and collaborators are not responsible about the contents of the generated voices by different individuals!