NodeJS Game Voice test client

This is a NodeJS example of recording from the microphone and streaming to DeepSpeech with voice activity detection, and sending commands via UART to HID input generator.

Based off nodejs_mic_vad_streaming from https://github.com/mozilla/DeepSpeech-examples/

Modified mic library to reduce stream chunk size and workaround for sox pipe bug.

Prerequisites:

1. The example utilized the mic NPM module which requires either sox (Windows/Mac) or arecord (Linux).

2. Download the pre-trained DeepSpeech english model (1.8GB):

wget https://github.com/mozilla/DeepSpeech/releases/download/v0.6.0/deepspeech-0.6.0-models.tar.gz
tar xvfz deepspeech-0.6.0-models.tar.gz

Install:

npm install

Run NodeJS server:

node start.js

Specify alternate DeepSpeech model path:

Use the DEEPSPEECH_MODEL environment variable to change models. Use the DEEPSPEECH_LM environment variable to change langauge model.

DEEPSPEECH_MODEL=~/dev/jaxcore/deepspeech-0.6.0-models/ node start.js