khanomtan-tts-v1.0

KhanomTan TTS v1.0

KhanomTan TTS (ขนมตาล) is an open-source Thai text-to-speech model that supports multilingual speakers such as Thai, English, and others.

KhanomTan TTS is a YourTTS model trained on multilingual languages that supports Thai. We use Thai speech corpora, TSync 1* and TSync 2* mbarnig/lb-de-fr-en-pt-12800-TTS-CORPUS to train the YourTTS model by using code from the 🐸 Coqui-TTS.

Config

We use Thai characters to the graphemes config to training the model and use the Speaker Encoder model from 🐸 Coqui-TTS.

Dataset

We use Tsync 1 and Tsync 2 corpora, which are not complete datasets, and then add these to mbarnig/lb-de-fr-en-pt-12800-TTS-CORPUS dataset.

Trained the model

We use the 🐸 Coqui-TTS multilingual VITS-model recipe (version 0.7.1 or the commit id is d46fbc240ccf21797d42ac26cb27eb0b9f8d31c4) for training the model, and we use the speaker encoder model from 🐸 Coqui-TTS then we release the best model to public access.

• Model cards: https://github.com/wannaphong/KhanomTan-TTS-v1.0
• Dataset (Tsync 1 and Tsync 2 only): https://huggingface.co/datasets/wannaphong/tsync1-2-yourtts
• GitHub: https://github.com/wannaphong/KhanomTan-TTS-v1.0

*Note: Those are not complete corpus. We can access the public corpus only.