Skip to content

README.md

Latest commit

 

History

History
157 lines (127 loc) · 5.51 KB

README.md

File metadata and controls

157 lines (127 loc) · 5.51 KB
Raw

Configuration Files

This directory contains JSON configuration files that control AI prompts and voice settings for the AudioNews service.

Files

ai_prompts.json

Controls all AI-related prompts and model settings.

Structure:

  • system_messages: Language-specific system messages for the AI (instructions for accessible content creation)
  • analysis_prompt: Configuration for news story analysis and categorization
    • template: The prompt template for analyzing headlines
    • system_instruction: Wrapper instruction emphasizing duplicate elimination
    • region_names: Language-specific region names (e.g., "UK", "French", "German")
  • synthesis_prompts: Language-specific prompts for generating audio content
    • Each language has a template field with placeholders for {theme}, {headlines}, and {previous_context}
    • {previous_context} enables context-aware generation to avoid repetition across themes
    • Each language has a max_words field (80 for most languages, 100 for BellaNews)
  • ai_model: Claude model configuration
    • name: Model identifier (e.g., "claude-sonnet-4-5-20250929")
    • analysis_max_tokens: Max tokens for story analysis (1500)
    • analysis_temperature: Temperature for analysis (0.1 for consistency)
    • synthesis_max_tokens: Max tokens for content synthesis (300)
    • synthesis_temperature: Temperature for synthesis (0.4 for creativity)

Supported Languages:

  • en_GB: English (UK) - Active
  • pl_PL: Polish (Poland) - Active
  • bella: BellaNews (Personalized business/finance) - Active

Disabled Languages (cost optimization):

  • en_GB_LON: English (London)
  • en_GB_LIV: English (Liverpool)
  • fr_FR: French (France)
  • de_DE: German (Germany)
  • es_ES: Spanish (Spain)
  • it_IT: Italian (Italy)
  • nl_NL: Dutch (Netherlands)

voice_config.json

Controls voice settings and TTS configuration.

Structure:

  • voices: One entry per language – each digest (en_GB, pl_PL, bella) uses its own voice so the three audio outputs sound distinct.
    • name: Edge TTS voice (e.g. "en-IE-EmilyNeural", "pl-PL-ZofiaNeural")
    • tts_provider: edge_tts | pocket_tts | elevenlabs
    • pocket_voice: Pocket TTS voice id (e.g. "alba") when using Pocket TTS
    • elevenlabs_voice_id: ElevenLabs voice id when using ElevenLabs (e.g. "EXAVITQu4vr4xnSDxMaL" = Rachel, "pNInz6obpgDQGcFmaJgB" = Adam)
    • display_name, language, gender, provider: metadata
  • tts_settings: Provider-specific options (Edge, Pocket, ElevenLabs)
    • edge_tts:
      • max_retries: Number of retry attempts (5)
      • initial_retry_delay: Initial delay in seconds (5)
      • retry_backoff_multiplier: Backoff multiplier for exponential delay (1.5)
      • force_ipv4: Force IPv4 connections for GitHub Actions compatibility (true)
      • ssl_verify: SSL certificate verification (true)
      • rate: Speech rate adjustment (default: "+10%")
        • Valid range: "-50%" to "+100%"
        • Examples: "+10%" (10% faster), "+20%" (20% faster), "0%" (normal), "-10%" (10% slower)
        • Recommended: "+10%" to "+15%" for optimal speech rate (120-150 WPM)
    • fallback:
      • enabled: Whether fallback is enabled (false)
      • provider: Fallback provider if Edge TTS fails ("google_tts")
  • audio_settings: Audio file settings
    • format: Audio format ("mp3")
    • quality: Quality level ("high")
    • estimated_words_per_second: Speech rate estimation (2.0)
    • min_file_size_kb: Minimum acceptable file size (100)

Usage

Configuration is loaded by the digest package when you run the main script:

python scripts/github_ai_news_digest.py --language en_GB

The digest package reads config/ai_prompts.json and config/voice_config.json and builds language configs (including sources and themes from digest/config_loader.py).

Editing Guidelines

Adding a New Language

  1. In ai_prompts.json:

    • Add system message to system_messages object
    • Add region name to analysis_prompt.region_names
    • Add synthesis prompt template to synthesis_prompts

  2. In voice_config.json:

    • Add voice configuration to voices object with appropriate Edge TTS (or other provider) voice

  3. In digest/config_loader.py:

    • Add the language to the templates dictionary with sources, themes, greeting, region_name, output_dir, audio_dir, service_name (see existing entries such as en_GB or pl_PL)

Changing AI Model

Update the ai_model section in ai_prompts.json:

{
  "ai_model": {
    "name": "claude-sonnet-4-5-20250929",
    "analysis_max_tokens": 1500,
    "analysis_temperature": 0.1,
    "synthesis_max_tokens": 300,
    "synthesis_temperature": 0.4
  }
}

Changing Voice

Update the voice name in voice_config.json:

{
  "voices": {
    "en_GB": {
      "name": "en-IE-EmilyNeural",
      ...
    }
  }
}

Available Edge TTS voices: https://speech.microsoft.com/portal/voicegallery

Adjusting Retry Logic

Update tts_settings.edge_tts in voice_config.json:

{
  "tts_settings": {
    "edge_tts": {
      "max_retries": 5,
      "initial_retry_delay": 5,
      "retry_backoff_multiplier": 1.5
    }
  }
}

Testing Changes

After editing configuration files, test that they load correctly:

python3 -c "from digest.config_loader import AI_PROMPTS_CONFIG, VOICE_CONFIG, LANGUAGE_CONFIGS; print('✅ Config loaded')"

Version Control

Always commit configuration changes with clear descriptions:

git add config/
git commit -m "🔧 Update AI prompts for better synthesis"