Hi, I have a retaurant bot with my custom training data. and i want to integrate this bot with mozilla voice frameworks stt and tts, and i dont want to use pre trained models for both tts and stt(mozilla deep speech) like sara bot does in the blog post How to build a voice assistant with open source Rasa and Mozilla tools. Any solution guys?
@Juste your help is required as you have written this How to build a voice assistant with open source Rasa and Mozilla tools blog as well
Hi @noman. You can actually train your own models form scratch using Mozilla if you like, but for that you will need quite a bit of data here you can find some details on that DeepSpeech/TRAINING.rst at master · mozilla/DeepSpeech · GitHub. Any reason why you don’t want to use the pretrained models?
1 Like
Thanks @Juste for the quick response. Actually my bot needs to support URDU language (probably urdu-speech to roman-urdu-text and roman-urdu-text to urdu-speech )as well but the pretrained models are specifically for ENGLISH language. How can i achieve this?
Oh, I see. I think it’s now possible to build the models for Urdu language (see this thread on Mozilla deep speech: https://github.com/mozilla/DeepSpeech/issues/634). You should be able to train your own model, but as I have mentioned, for that you will need the training data - audio samples + text transcripts.
An easy alternative (not open source though) is to use Google Speech API. Based on the documentation I can see that they support Urdu language (https://cloud.google.com/speech-to-text/docs/languages)
1 Like
Great i will look into this. Thanks for directing me and sorry for taking your precious time.
Let us know how the project goes!