Translation

Automatically detects your input language and translates both your message to English (for the AI) and the AI's response back to your detected language.

Specify the exact language you want responses in, regardless of your input language. This gives you precise control over the output language.

Auto-Detection in Voice Agent

Voice Agent automatically detects your spoken language and responds in the same language:

1import { io } from 'socket.io-client';
2
3// Connect to Voice Agent
4const socket = io('wss://api.selamgpt.com/v1/audio/agent', {
5  query: {
6    token: 'your-jwt-token',
7    api_key: 'your-api-key',
8    voice: 'mekdes'  // Amharic voice
9  },
10  transports: ['websocket']
11});
12
13socket.on('connected', () => {
14  // Start session without specifying language
15  // Voice Agent will auto-detect from speech
16  socket.emit('start_session', {
17    voice: 'mekdes',
18    instructions: 'You are a helpful assistant.'
19    // No language parameter - auto-detect mode
20  });
21});
22
23// Speak in Amharic, get responses in Amharic
24// Speak in English, get responses in English

Target Language in Voice Agent

Specify a target language to always receive responses in that language:

1import { io } from 'socket.io-client';
2import PageFeedback from '@/components/docs/PageFeedback';
3
4const socket = io('wss://api.selamgpt.com/v1/audio/agent', {
5  query: {
6    token: 'your-jwt-token',
7    api_key: 'your-api-key',
8    voice: 'ubax'  // Somali voice
9  },
10  transports: ['websocket']
11});
12
13socket.on('connected', () => {
14  // Start session with target language
15  socket.emit('start_session', {
16    voice: 'ubax',
17    language: 'so',  // Always respond in Somali
18    instructions: 'You are a helpful assistant.'
19  });
20});
21
22// Speak in any language, get responses in Somali

Choose the Right Mode

• •Use auto mode when you want the simplest experience and will consistently use one language
• •Use target mode when you need precise control over output language
• •Use default mode (no translation) when working in English to avoid translation overhead

Handle Errors Gracefully

• •Always check for translation errors and provide fallback behavior
• •Implement retry logic for translation service unavailability (503 errors)
• •Handle language detection failures by falling back to target mode

Monitor Performance

• •Use translation_metadata to track translation performance
• •Translation adds ~200-500ms overhead - factor this into your application design
• •Use streaming for long responses to improve perceived latency

Ensure Proper Encoding

• •Always use UTF-8 encoding for proper Unicode support
• •Set Content-Type: application/json; charset=utf-8 header
• •Verify your terminal/console supports UTF-8 for Ethiopian scripts

Test with Multiple Languages

• •Test your integration with all supported Ethiopian languages
• •Verify proper Unicode handling for each language's script
• •Test edge cases like mixed-language input and special characters

Optimize Input Length

• •Language detection works best with messages of at least 10-20 characters
• •For very short messages, consider using target mode instead of auto mode
• •Provide context in longer messages for better translation quality