Speak SDK Patterns

Overview

Production patterns for Speak language learning API: conversation sessions, pronunciation assessment, audio preprocessing, and batch operations.

Prerequisites

• Completed

  speak-install-auth

  setup
• Valid API credentials configured
• ffmpeg installed for audio processing

Instructions

Pattern 1: Conversation Session Manager

class ConversationManager {
  private client: SpeakClient;
  private sessions: Map<string, SessionState> = new Map();

  async startLesson(language: string, scenario: string, level: string) {
    const session = await this.client.startConversation({
      scenario, language, level, nativeLanguage: 'en',
    });
    this.sessions.set(session.id, {
      turns: [], startTime: Date.now(), language,
    });
    return session;
  }

  async submitResponse(sessionId: string, audioPath: string) {
    const turn = await this.client.sendTurn(sessionId, { audioPath });
    this.sessions.get(sessionId)?.turns.push(turn);
    return turn;
  }

  async endAndReport(sessionId: string) {
    const summary = await this.client.endSession(sessionId);
    const state = this.sessions.get(sessionId)!;
    return {
      ...summary,
      duration: (Date.now() - state.startTime) / 1000,
      totalTurns: state.turns.length,
      avgPronunciation: state.turns.reduce((s, t) =>
        s + (t.pronunciationScore || 0), 0) / state.turns.length,
    };
  }
}

Pattern 2: Audio Preprocessor

import { execSync } from 'child_process';

function preprocessAudio(inputPath: string): string {
  const outputPath = inputPath.replace(/\.[^.]+$/, '.processed.wav');
  // Convert to WAV 16kHz mono PCM — required by Speak API
  execSync(
    `ffmpeg -y -i "${inputPath}" -ar 16000 -ac 1 -c:a pcm_s16le "${outputPath}"`,
    { stdio: 'pipe' }
  );
  return outputPath;
}

Pattern 3: Retry with Backoff

async function withRetry<T>(fn: () => Promise<T>, maxRetries = 3): Promise<T> {
  for (let i = 0; i < maxRetries; i++) {
    try {
      return await fn();
    } catch (err: any) {
      if (err.response?.status === 429 && i < maxRetries - 1) {
        const wait = parseInt(err.response.headers['retry-after'] || '5');
        await new Promise(r => setTimeout(r, wait * 1000));
        continue;
      }
      throw err;
    }
  }
  throw new Error('Max retries exceeded');
}

Pattern 4: Progress Tracker

class LearningProgress {
  private history: SessionSummary[] = [];

  addSession(summary: SessionSummary) {
    this.history.push(summary);
  }

  getReport() {
    const recent = this.history.slice(-10);
    return {
      totalSessions: this.history.length,
      avgPronunciation: recent.reduce((s, h) => s + h.avgPronunciationScore, 0) / recent.length,
      totalMinutes: this.history.reduce((s, h) => s + h.durationMinutes, 0),
      vocabularyLearned: [...new Set(this.history.flatMap(h => h.newWords))].length,
    };
  }
}

Output

• Patterns implementation complete
• Speak API integration verified
• Production-ready patterns applied

Error Handling

Error	Cause	Solution
401 Unauthorized	Invalid API key	Verify SPEAK_API_KEY environment variable
429 Rate Limited	Too many requests	Wait Retry-After seconds, use backoff
Audio format error	Wrong codec/sample rate	Convert to WAV 16kHz mono with ffmpeg
Session expired	Timeout after 30 min	Start a new conversation session

Resources

• Speak Website
• OpenAI Realtime API
• Speak GPT-4 Blog

Next Steps

See

speak-prod-checklist

for production readiness.

Examples

Basic: Apply sdk patterns with default configuration for a standard Speak integration.

Advanced: Customize for production with error recovery, monitoring, and team-specific requirements.