Syntethic voices generation in Hindi

Hi,

I want to create a dataset with synthetic voices in Hindi Language, I suppose tacotron and wavenet are trained on English. Any idea how can I go about collecting data to train a production grade speech to text engine?

You should look into the Common Voice Project.

Could you explain a little more about what you’re trying as your overall objective?

You seem to be mixing comments that interchangeably imply an interest in TTS and STT and you’ve posted this in the STT forum in spite of a title that looks more appropriate for TTS. I appreciate they’re not completely disconnected but would be good to understand if you’re clear on your own understanding first (no offence intended :slightly_smiling_face: )

Hi @nmstoker

The OP wants to train a speech-to-text model (STT). However, he needs to generate data to train it. He has the text files and wants to generate the speech files from it (i.e. do TTS) :slight_smile:

Thanks @xhtm - I guessed as much but don’t know why they didn’t just say that more clearly.

Also if they have (or can conceivably get) the data to produce a TTS but ultimately want to produce STT why not start with at least trying to use that data for transfer learning / fine tuning with the STT model?

I think they only have the text files, and are looking for a software that can produce speech for them. So, right now they can’t really do STT.