인터페이스는 현지화되어 있으며, Skill 소개와 설명 문서는 선택한 언어로 AI가 자동 번역합니다. 코드·명령어·경로·전문 용어는 원문을 유지하며 원작자 문서가 우선합니다.
Skills 둘러보기
개발 및 엔지니어링 · SKILL.md

speak-hello-world

speak-hello-world

Create your first Speak AI tutoring session with pronunciation feedback. Use when starting a new Speak integration, testing your setup, or learning basic language learning API patterns. Trigger with phrases like "speak hello world", "speak example", "speak quick start", "first speak lesson".

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill 문서

GitHub

name: speak-hello-world description: 'Create your first Speak AI tutoring session with pronunciation feedback.

Use when starting a new Speak integration, testing your setup,

or learning basic language learning API patterns.

Trigger with phrases like "speak hello world", "speak example",

"speak quick start", "first speak lesson".

' allowed-tools: Read, Write, Edit, Bash(npm:), Bash(node:) version: 1.14.0 license: MIT author: Jeremy Longshore jeremy@intentsolutions.io tags:

  • saas
  • speak
  • api
  • testing compatibility: Designed for Claude Code

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

ErrorCauseSolution
Session timeoutExceeded max durationStart a new session
Audio format invalidWrong codec or sample rateConvert to WAV 16kHz mono
Language not supportedInvalid language codeUse supported codes (es, ko, ja, fr, de)
Low pronunciation scoreBackground noiseRecord in a quiet environment
Rate limit exceededToo many requestsWait and retry with backoff

Resources

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).

저장한 Skills