IBM's speech-to-text demo: just add credentials and pray

What it does This is a sample Node.js application demonstrating IBM Watson’s Speech to Text service. It accepts audio input, streams it to IBM’s cloud API via WebSocket, and returns live transcriptions that self-correct as more speech arrives. The repo includes local setup instructions and a Cloud Foundry deployment path for IBM Cloud.

The interesting bit The WebSocket interface is the default path here, not an afterthought — the demo leans into streaming transcription rather than batch upload. That’s rarer in sample code than it should be, though the README treats it as barely worth mentioning.

Key highlights

• Live transcription with incremental correction as audio continues
• WebSocket-first, REST fallback
• One-command local run (npm start on port 3000)
• Cloud Foundry manifest included for IBM Cloud deployment
• Apache 2.0 licensed

Caveats

• Requires IBM Cloud account, CLI tooling, and service provisioning before you can even run it locally
• The README’s example .env file contains what looks like a real API key (likely rotated, but still: don’t copy-paste blindly)
• No discussion of audio format requirements, rate limits, or pricing tiers

Verdict Grab this if you’re already committed to IBM Cloud and need a working baseline to hack on. Skip it if you’re comparing speech APIs or want something that runs without cloud vendor lock-in.