Voice Input for Chatbot: Noise Handling & Custom Vocabulary

Feature Request: Voice Input for Chatbot

Overview

Add voice input capability to the chatbot that can handle industrial/factory environments with noise and recognize domain-specific vocabulary.

Requirements

1. Noise Robustness

• Handle background noise common in factory environments (machinery, alarms, ventilation)
• Support voice activity detection (VAD) to distinguish speech from ambient noise
• Consider noise suppression/cancellation preprocessing

2. Custom Vocabulary Support

• Recognize domain-specific terms: equipment names, part numbers, procedure names
• Examples from XFab context:
• Equipment: "Novellus SPEED", "Sequel", "Altus", "ESC", "loadlock"
• Parts: "shower", "dome", "slit", "gas ring", "heater bloc"
• Procedures: "handoff", "recovery", "crushball test"
• Support for French technical terms (multilingual factory environment)

Technical Considerations

Speech-to-Text Options

| Option | Pros | Cons |
|--------|------|------|
| Google Cloud Speech-to-Text | Custom vocabulary, noise robustness, phrase hints | Cost per minute |
| Whisper (OpenAI) | High accuracy, multilingual | No custom vocabulary boosting |
| Deepgram | Real-time, custom vocabulary, noise handling | Third-party dependency |
| AssemblyAI | Good noise handling, custom spelling | Cost |

Recommended Approach

1. Primary: Google Cloud Speech-to-Text with phrase hints for custom vocabulary
2. Preprocessing: WebRTC-based noise suppression in browser before sending audio
3. Fallback: Allow text input when voice fails

Implementation Notes

• Frontend: Web Audio API for capture + noise suppression
• Consider push-to-talk vs continuous listening modes
• Cache custom vocabulary per company (equipment names from knowledge graph)

Acceptance Criteria

• [ ] Voice input button in chat interface
• [ ] Handles 70+ dB ambient noise environments
• [ ] 90%+ accuracy on domain-specific terms
• [ ] Works in Chrome, Safari, Firefox
• [ ] Graceful fallback when microphone unavailable

Related

• Chat module: apps/api/app/services/chat_module/
• Frontend chat: apps/web/src/components/chat/