while creating voice assistant in rasa, is it possible to train my own stt model (to identify only specific words) or should i use existing deepseech models.
Training a stt model takes a lot of time, data and computing power. So I recommend you to use a pretrained model.
If you can’t find a model for your needs, search for one that allows fine tuning (the training of only the last layers of the network) so you can add your own data to the model in a faster way.
- But what if i want to use voice assistance for other language like hindi etc. 2)Can i integrate other models that i trained using kaldi or wav2letter instead of deepspeech.
Yes. You just have to create a connector to rasa.