Speak Hello World

Overview

Create your first AI tutoring session with Speak. Demonstrates conversation practice, pronunciation assessment, and real-time feedback using GPT-4o-powered tutoring.

Prerequisites

• Completed

  speak-install-auth

  setup
• Valid API credentials configured
• Microphone access (optional for testing)

Instructions

Step 1: Start a Conversation Session

import { SpeakClient } from '@speak/language-sdk';

const client = new SpeakClient({
  apiKey: process.env.SPEAK_API_KEY!,
  appId: process.env.SPEAK_APP_ID!,
  language: 'es',
});

// Start a beginner Spanish lesson
const session = await client.startConversation({
  scenario: 'greetings',
  language: 'es',
  level: 'beginner',
  nativeLanguage: 'en',
});

console.log('Session ID:', session.id);
console.log('AI Tutor:', session.firstPrompt.text);
// Output: "Hola! Bienvenido a tu leccion de espanol. Como te llamas?"
console.log('Audio URL:', session.firstPrompt.audioUrl);

Step 2: Send a Student Response

// Submit text response (or audio file for pronunciation scoring)
const turn = await client.sendTurn(session.id, {
  text: 'Hola, me llamo Juan. Mucho gusto.',
  // Or: audioPath: './recordings/response.wav'
});

console.log('Tutor response:', turn.tutorText);
console.log('Pronunciation score:', turn.pronunciationScore); // 0-100
console.log('Grammar corrections:', turn.corrections);
// Output: [{original: "me llamo", suggestion: null, correct: true}]
console.log('Vocabulary notes:', turn.vocabularyNotes);

Step 3: Pronunciation Assessment

// Assess pronunciation of a specific phrase
const assessment = await client.assessPronunciation({
  audioPath: './recordings/hola-como-estas.wav',
  targetText: 'Hola, como estas?',
  language: 'es',
  detailLevel: 'phoneme', // 'word' or 'phoneme'
});

console.log(`Overall score: ${assessment.score}/100`);
for (const word of assessment.words) {
  console.log(`  "${word.text}": ${word.score}/100`);
  if (word.phonemes) {
    for (const p of word.phonemes.filter(p => p.score < 70)) {
      console.log(`    Weak phoneme: ${p.symbol} (${p.score}) - ${p.suggestion}`);
    }
  }
}

Step 4: End Session and Review

const summary = await client.endSession(session.id);
console.log('Session Summary:');
console.log(`  Duration: ${summary.durationMinutes} min`);
console.log(`  Turns: ${summary.totalTurns}`);
console.log(`  Pronunciation: ${summary.avgPronunciationScore}/100`);
console.log(`  Grammar: ${summary.grammarAccuracy}%`);
console.log(`  New vocabulary: ${summary.newWords.join(', ')}`);

Output

• Working conversation session with AI tutor
• Pronunciation assessment with phoneme-level feedback
• Session summary with learning metrics
• Console output showing scores and corrections

Error Handling

Error	Cause	Solution
Session timeout	Exceeded max duration	Start a new session
Audio format invalid	Wrong codec or sample rate	Convert to WAV 16kHz mono
Language not supported	Invalid language code	Use supported codes (es, ko, ja, fr, de)
Low pronunciation score	Background noise	Record in a quiet environment
Rate limit exceeded	Too many requests	Wait and retry with backoff

Resources

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

Next Steps

Proceed to

speak-local-dev-loop

for development workflow setup.

Examples

Text-only test: Skip audio and use text responses to test the conversation flow before integrating microphone input.

Multi-language: Start sessions in different languages by changing the

language

parameter to

ko

(Korean),

ja

(Japanese), or

fr

(French).