Language Support¶

Voice Transcriber supports multiple languages with strong language enforcement to prevent unwanted language switching during transcription.

Supported Languages¶

Configuring Language¶

Edit your configuration file:

nano ~/.config/voice-transcriber/config.json

Set the language field:

{
  "language": "fr",
  "formatterEnabled": true,
  "transcription": {
    "backend": "openai",
    "openai": {
      "apiKey": "sk-..."
    }
  }
}

Restart the application for changes to take effect.

Language Enforcement¶

Voice Transcriber uses strong language-specific prompts to prevent Whisper from switching languages mid-transcription.

How It Works¶

When you set a language (e.g., French):

1. Transcription prompt explicitly instructs Whisper:

   This is a French audio recording. Transcribe the entire audio in French only.
   Do NOT switch to English or translate. Keep all content in French.
   

2. Formatting prompt maintains the language:

   Format this French text with proper grammar and punctuation.
   Keep the text in French. Do not translate.
   

This dual-layer enforcement ensures your transcriptions stay in the configured language.

Examples by Language¶

English¶

Recording: "Hello, this is a test of the transcription system"

Result: "Hello, this is a test of the transcription system."

French¶

Recording: "Bonjour, ceci est un test du système de transcription"

Result: "Bonjour, ceci est un test du système de transcription."

Spanish¶

Recording: "Hola, esta es una prueba del sistema de transcripción"

Result: "Hola, esta es una prueba del sistema de transcripción."

Mixed-Language Content¶

For recordings with technical English terms in French sentences (common in tech contexts):

Configuration:

{
  "language": "fr",
  "transcriptionPrompt": "Transcribe this French audio. Keep technical English terms but preserve French grammar.",
  "transcription": {
    "backend": "openai",
    "openai": {
      "apiKey": "sk-..."
    }
  }
}

Recording: "J'utilise React avec TypeScript pour développer l'application"

Result: "J'utilise React avec TypeScript pour développer l'application."

Troubleshooting Language Issues¶

French Transcription Switches to English¶

Problem: Long French recordings switch to English mid-way

Solution: Ensure language is set explicitly:

{
  "language": "fr",
  "transcription": {
    "backend": "openai",
    "openai": {
      "apiKey": "sk-..."
    }
  }
}

Restart the application after config changes.

Technical Terms Not Recognized¶

Problem: Technical English terms in French speech are poorly transcribed

Solution: Use custom transcription prompt:

{
  "language": "fr",
  "transcriptionPrompt": "Transcribe this French technical audio. Keep common English technical terms as-is.",
  "transcription": {
    "backend": "openai",
    "openai": {
      "apiKey": "sk-..."
    }
  }
}

Accent Not Recognized Well¶

Problem: Strong accent reduces accuracy

Solutions: - Speak slightly slower and more clearly - Reduce background noise for better recognition - Try self-hosted Speaches with different models (see Speaches Integration)

Language-Specific Tips¶

French¶

• Accents (é, è, à, ç) are generally recognized well
• Homophones (ou/où, a/à) may need manual correction
• Long compound sentences benefit from formatting

Spanish¶

• Accents and ñ are well recognized
• Regional variations (Spain vs Latin America) work well
• Inverted question marks may need manual addition

German¶

• Umlauts (ä, ö, ü, ß) are recognized accurately
• Compound words are usually transcribed correctly
• Formal/informal you (Sie/du) preserved correctly

Italian¶

• Accents (à, è, ì, ò, ù) are well recognized
• Double consonants are usually correct
• Regional variations work well

Performance by Language¶

All supported languages have similar processing times:

Adding More Languages¶

While only 5 languages have strong enforcement prompts, Whisper supports many more:

Other supported codes: ja, zh, pt, ru, ko, ar, nl, pl, tr, etc.

To use unsupported language:

{
  "language": "pt",
  "transcriptionPrompt": "Transcribe this Portuguese audio completely in Portuguese. Do not switch languages.",
  "transcription": {
    "backend": "openai",
    "openai": {
      "apiKey": "sk-..."
    }
  }
}

Next Steps¶

• Basic Usage - Learn recording workflow
• Configuration Guide - Custom prompts and settings
• Transcription Backends - OpenAI vs Speaches