Build a Voice Command Assistant with Node.js on Raspberry Pi

Build a voice-controlled assistant on a Raspberry Pi 3 using Node.js, PocketSphinx for speech recognition, and Node-RED for command orchestration.

Architecture

Three components communicate via MQTT:

• shirka_ears: Listens to microphone, publishes voice commands
• shirka_voice: Receives messages, converts to speech, plays audio
• Node-RED: Routes commands and responses between components

The Ears

Speech recognition using PocketSphinx:

npm install pocketsphinx

Listen for commands and publish to MQTT:

const mqtt = require('mqtt');
const ps = require('pocketsphinx');

const client = mqtt.connect('mqtt://localhost');

ps.listen()
  .on('data', (command) => {
    client.publish('shirka/commands', command);
  });

The Voice

Text-to-speech pipeline using pico2wave and omxplayer:

sudo apt-get install libttspico-utils omxplayer

Subscribe to MQTT, generate speech, play audio:

const mqtt = require('mqtt');
const { exec } = require('child_process');

const client = mqtt.connect('mqtt://localhost');

client.subscribe('shirka/responses');

client.on('message', (topic, message) => {
  const text = message.toString();
  exec(`pico2wave -w response.wav "${text}" && omxplayer response.wav`);
});

Node-RED Flow

Install Node-RED and MQTT broker:

sudo apt-get install mosquitto mosquitto-clients
npm install -g node-red

Create a flow that:

1. Subscribes to shirka/commands
2. Processes commands
3. Publishes responses to shirka/responses

Code

Full implementation: github.com/bustroker/bustroker.shirka