Text front-end
Numbers are expanded to words, sentences are segmented, and English words are transliterated into the native script.
Natural speech for the next billion users.
Type in any of 13 Indian languages and hear it spoken back in a male or female voice. Open-source models, running on your own server.
Recent
Thirteen languages, ten scripts.
Dravidian, Indo-Aryan and Tibeto-Burman languages, each with jointly trained male and female voices.
How a sentence becomes speech
Acoustic model
A non-autoregressive model predicts pitch and duration per character, then renders a mel-spectrogram for the chosen speaker.
Neural vocoder
The spectrogram is turned into a waveform, denoised, and returned to your browser as WAV.