Speak local dev loop

Speak Local Dev Loop

Overview

Set up a fast local development workflow for Speak language learning integrations. Includes mock tutor responses for offline development, audio test fixtures, and a debug mode for scoring analysis.

Prerequisites

• Completed speak-install-auth setup
• Node.js 18+ with npm/pnpm
• ffmpeg installed (for audio conversion)

Instructions

Step 1: Project Structure

speak-app/
  src/
    speak/client.ts       # Speak API client
    speak/tutor.ts        # AI tutor conversation manager
    speak/pronunciation.ts # Pronunciation assessment
    speak/audio.ts        # Audio recording and preprocessing
  tests/
    mocks/                # Mock responses for offline dev
    fixtures/             # Sample audio files for testing
    unit/                 # Unit tests with mocked API
    integration/          # Integration tests (needs API key)
  .env.development        # Dev credentials
  .env.test              # Test credentials (mock mode)

Step 2: Mock Tutor for Offline Development

// tests/mocks/mock-speak-client.ts
export class MockSpeakClient {
  async startConversation(config: any) {
    return {
      id: 'mock-session-123',
      firstPrompt: {
        text: 'Hola! Bienvenido. Como te llamas?',
        audioUrl: null,
      },
    };
  }

  async sendTurn(sessionId: string, input: any) {
    return {
      tutorText: 'Muy bien! Tu pronunciacion es buena.',
      pronunciationScore: 85,
      corrections: [],
      vocabularyNotes: ['llamo = I call myself'],
    };
  }

  async assessPronunciation(config: any) {
    return {
      score: 82,
      words: [
        { text: 'Hola', score: 95, phonemes: [] },
        { text: 'como', score: 78, phonemes: [
          { symbol: 'o', score: 65, suggestion: 'Round lips more' }
        ]},
      ],
    };
  }
}

Step 3: Audio Test Fixtures

# Generate test audio files using text-to-speech
# macOS:
say -v "Paulina" "Hola, como estas" -o tests/fixtures/hola-es.wav
# Linux:
espeak -v es "Hola, como estas" -w tests/fixtures/hola-es.wav

# Convert to required format
ffmpeg -i tests/fixtures/hola-es.wav -ar 16000 -ac 1 tests/fixtures/hola-es-16k.wav

Step 4: Development Scripts

{
  "scripts": {
    "dev": "tsx watch src/index.ts",
    "dev:mock": "SPEAK_MOCK_MODE=true tsx watch src/index.ts",
    "test": "vitest run",
    "test:watch": "vitest",
    "lesson:test": "tsx src/test-lesson.ts"
  }
}

Step 5: Debug Mode

// Enable detailed scoring output
const DEBUG = process.env.SPEAK_DEBUG === 'true';

function logAssessment(result: PronunciationResult) {
  if (!DEBUG) return;
  console.log('=== Pronunciation Debug ===');
  for (const word of result.words) {
    const flag = word.score < 70 ? 'WEAK' : 'OK';
    console.log(`  [${flag}] "${word.text}": ${word.score}/100`);
    for (const p of word.phonemes || []) {
      if (p.score < 70) {
        console.log(`    phoneme "${p.symbol}": ${p.score} - ${p.suggestion}`);
      }
    }
  }
}

Output

• Project structure with client, tests, and mocks
• Mock tutor client for offline development
• Audio test fixtures generated with TTS
• Debug mode for pronunciation analysis
• Development scripts for fast iteration

Error Handling

Error	Cause	Solution
ffmpeg not found	Not installed	Install via brew install ffmpeg or apt install ffmpeg
Mock mode not activating	Env var not set	Check SPEAK_MOCK_MODE=true
Audio file too large	Long recording	Trim to 30 seconds max for testing
Test timeout	Slow API response	Use mock mode for unit tests

Resources

• Vitest
• ffmpeg
• Web Speech API

Next Steps

See speak-sdk-patterns for production-ready code patterns.

Examples

Mock-first development: Run npm run dev:mock to iterate on UI and conversation flow without API calls. Switch to real API for integration testing.

Audio quality testing: Record samples at different quality levels and compare pronunciation scores to calibrate your recording pipeline.