This video quickly shows Rasa X the web based tool for working with chatbots and then runs through a demo I've recently updated to work with Rasa 2.0 and then hooked up with a speech project I put together called Simple Speech Loop so I can talk to and hear responses from the bot.
The bot is a demo that works with a table of facts about UK monarchs and can handle a variety of questions about that topic, which it translates into queries against the table and then feeds that back to the user.
Input is via the microphone, feeding local speech recognition (Wav2Vec 2.0) with output speech using a TTS model I trained on public domain speech.
One of the nice capabilities demonstrated here is that the chatbot, speech recognition and text to speech all run entirely locally (ie nothing sent off to the cloud) and so could be run entirely on prem and/or on private cloud infrastructure.
Links:
Rasa X:
Bot: (at time of posting the code here is not the updated 2.0 version)
Simple Speech Loop: (video swaps in Wav2Vec for DeepSpeech on the ASR side)
Wav2Vec 2.0:
TTS: (video model is EK1 which isn't currently downloadable but was trained using the code in this repo against public domain speech from LibriVox)
