Voice AI Backend with Express
Budget: ₹12,500 – ₹37,500 INR
I’m building a web-based voice assistant that can listen, reason and speak back to the user, and I need the server-side logic developed in Node.js using Express.js.
Core requirements
• Accept live or recorded audio, run speech-to-text, pass the transcript through my business logic (I’ll supply the rules/LLM calls), then return synthesized speech.
• Provide an endpoint that can also place or receive phone calls, stream the audio both ways, and run the same STT / TTS pipeline in real time.
• Expose clean JSON APIs plus a WebSocket channel so the front-end can show partial transcripts and intermediate thinking steps.
• All configuration (API keys for Google Cloud, Amazon Polly, Twilio, etc.) must be environment-driven and easy to swap.
• Unit-tested code, concise README and a short screencast that proves the flow from browser to call and back.
Acceptance criteria
1. Docker-compose up builds and starts everything, including any Redis/queue you add.
2. Hitting /health returns 200.
3. Posting an audio file to /voice returns a playable TTS response within 3 s for a 5-second clip.
4. During a live call the latency between spoken word and assistant reply stays under 1 s on average.
Let me know which STT/TTS stack you favour and any past projects where you streamed audio through Express; that context will help me move fast.
Core requirements
• Accept live or recorded audio, run speech-to-text, pass the transcript through my business logic (I’ll supply the rules/LLM calls), then return synthesized speech.
• Provide an endpoint that can also place or receive phone calls, stream the audio both ways, and run the same STT / TTS pipeline in real time.
• Expose clean JSON APIs plus a WebSocket channel so the front-end can show partial transcripts and intermediate thinking steps.
• All configuration (API keys for Google Cloud, Amazon Polly, Twilio, etc.) must be environment-driven and easy to swap.
• Unit-tested code, concise README and a short screencast that proves the flow from browser to call and back.
Acceptance criteria
1. Docker-compose up builds and starts everything, including any Redis/queue you add.
2. Hitting /health returns 200.
3. Posting an audio file to /voice returns a playable TTS response within 3 s for a 5-second clip.
4. During a live call the latency between spoken word and assistant reply stays under 1 s on average.
Let me know which STT/TTS stack you favour and any past projects where you streamed audio through Express; that context will help me move fast.