ScaleReach's Speech Service

Handles speech synthesis and recognition within ScaleReach with multilanguage support

• Recognition: Uses websockets to streams raw audio data in opus codecs in mpeg containers for live transcription
• Synthesis: Uses simple HTTP requests to obtain streamable audio data in aac codecs

Startup

To startup a development server,

npm i
npm start

Alternatively, you may use pnpm as a drop-in replacement for npm

Lightweight server with no build processes

/src/config.js

Configuration for frontend

• .invokeOrigin: (string) URL address of ScaleReach's frontend (e.g. "http://localhost:8000", no trailing forward slash), used to establish CORS policy for web sockets establishment

.env

• DEEPGRAM_API_KEY: (string) DeepGram's supplied API key
• PORT: (number) port number the Speech Service server will run on (defined in /Dockerfile)

Dependencies

• express.js: NodeJS web framework
• socket.IO: NodeJS web socket library
• Deepgram JS SDK: JS SDK to interact with Deepgram's API

Deploy

Ensure .env file exists within /src folder (e.g. /src/.env)

1. Upload certificates (.pem and .key) into /certs (./certs/scalereach.team.pem and ./certs/scalereach.team.key)
2. docker build -t transcriber .
3. docker run -p 6733:6733 --env-file ./src/.env transcriber