EasyTranslator Documentation

This directory contains locally cached documentation for the libraries and APIs used in the EasyTranslator project.

Contents

Web Speech API

Location: /docs/web-speech-api/

Browser-native speech recognition and synthesis API.

Files:

• README.md - Overview and quick start guide
• API.md - Complete API reference with all methods, properties, and events
• USAGE.md - Detailed usage guide with best practices and patterns
• examples/ - Working code examples
  • basic-recognition.js - Simple speech recognition
  • continuous-recognition.js - Continuous recognition with interim results
  • on-device-recognition.js - On-device processing with language pack management
  • contextual-biasing.js - Domain-specific recognition improvement
  • speech-synthesis.js - Text-to-speech with voice selection
• version.txt - Version info and compatibility notes

Key Features:

• Speech recognition (audio to text)
• Speech synthesis (text to audio)
• On-device processing (experimental)
• Contextual biasing
• Multilingual support

Browser Compatibility: Requires vendor prefixes (webkitSpeechRecognition)

---

Mistral Voxtral

Location: /docs/mistral-voxtral/

Mistral AI's frontier speech understanding models for transcription and audio Q&A.

Files:

• README.md - Overview, model variants, and quick start
• API.md - Complete API reference with endpoints and parameters
• examples/ - Integration examples
  • basic-transcription.js - Simple audio transcription
  • transcription-with-timestamps.js - Segment-level timestamps for subtitles
  • audio-chat.js - Q&A and summarization from audio
  • function-calling.js - Voice commands triggering backend functions
  • vue-composable.ts - Vue 3 composable for EasyTranslator integration
  • edge-function.ts - Serverless function to hide API key
• version.txt - Version info, pricing, and capabilities

Model Variants:

• Voxtral (24B) - Production scale (voxtral-small-latest)
• Voxtral Mini (3B) - Edge deployment (voxtral-mini-latest)
• Voxtral Mini Transcribe - API-optimized transcription

Key Features:

• 32k token context (30+ minutes audio)
• Built-in Q&A and summarization
• Native multilingual support (auto-detection)
• Function-calling from voice
• $0.001 per minute pricing

API Endpoints:

• POST /v1/audio/transcriptions - Transcription only
• POST /v1/chat/completions - Chat with audio input

---

Quick Reference

When to Use Which?

Use Web Speech API when:

• Need browser-native recognition (no API costs)
• Building voice commands for UI
• Want text-to-speech synthesis
• Privacy is critical (can run on-device)
• Working offline

Limitations:

• Server-based in Chrome (requires network)
• Less accurate than Voxtral
• No built-in translation
• Variable browser support

Use Mistral Voxtral when:

• Need high-accuracy transcription
• Want to ask questions about audio
• Building translation features
• Need function-calling from voice
• Working with long-form audio (up to 30 min)

Limitations:

• Requires API key and network
• Costs $0.001 per minute
• Not available offline

---

EasyTranslator Integration

Current Implementation

EasyTranslator currently uses Voxtral Mini for transcription:

// In useAudioRecorder.ts
const formData = new FormData()
formData.append('file', audioBlob, 'recording.webm')
formData.append('model', 'voxtral-mini')  // ← Current model

const response = await fetch('https://api.mistral.ai/v1/audio/transcriptions', {
  method: 'POST',
  headers: { 'x-api-key': apiKey },
  body: formData,
})

Potential Enhancements

1. Add Web Speech API as Fallback

   • Use for quick, local recognition
   • Fall back to Voxtral for accuracy
   • Enable offline mode

2. Implement Audio Q&A

   • Let users ask questions about conversations
   • Summarize long recordings
   • Extract action items from meetings

3. Enable Function Calling

   • Voice commands: "Translate to French"
   • Auto-detect: "Send this to my email"
   • Smart actions: "Save this conversation"

4. Add Timestamp Support

   • Generate subtitles from transcriptions
   • Enable seeking in long recordings
   • Create conversation chapters

5. Optimize with Edge Functions

   • Hide API key from frontend
   • Add rate limiting
   • Implement caching

---

Documentation Maintenance

Last Updated: 2025-11-26

Update Schedule:

• Check for API changes monthly
• Update examples when EasyTranslator architecture changes
• Add new examples based on user needs

How to Update:

1. Visit source URLs listed in each version.txt
2. Check for API changes or new features
3. Update relevant markdown files
4. Add new examples if needed
5. Update this README with changes

---

External Links

Web Speech API

• MDN: https://developer.mozilla.org/en-US/docs/Web/API/Web_Speech_API
• Spec: https://wicg.github.io/speech-api/

Mistral Voxtral

• Announcement: https://mistral.ai/news/voxtral
• API Docs: https://docs.mistral.ai/capabilities/audio_transcription
• Pricing: https://mistral.ai/pricing

---

Contributing

When adding new documentation:

1. Follow the existing structure
2. Include working code examples
3. Document browser/API compatibility
4. Update this README
5. Add version info to version.txt