Active Speech Recognition (ASR)

This repository implements an Active Speech Recognition. This allows users to speak into their mic and publish the transcribed text via a ZMQ socket.

ASR is primarily dependent on the following:

• Sounddevice: Microphone selection and audio input stream.
• Silero-VAD: Used to detect voice activity and determine end of utterance (EOU).
• WhisperX: Audio -> text transcription.

How to use:

1. Install the requirements by running the following pip3 install -r requirements.txt
2. Configure ASR settings.
   • Main configuration settings are located in config/config.yaml
     • To use another configuration file, pass the file path using --config
   • Override configuration settings using flags. For more information run python3 main.py --h
3. Main file is located at ASR/main.py. Run this file using python3 main.py
4. Begin speaking into the selected microphone.
5. If you need to check if the text is being published, run Examples/example_sub.py

Future Plans:

• Allow configurable VAD and transcription models.
• Create interface for microphone selection/settings.
• Enable different data transportion architectures beyond standard PUB/SUB.