Common audio formats
Description
Audio files are processed to create journal entries with automatic transcription and AI-generated titles. They support coordinate correlation through GPS track matching.
Supported Formats
All audio formats supported by pydub.AudioSegment, including:
- WAV (.wav)
- MP3 (.mp3)
- FLAC (.flac)
- M4A (.m4a)
- OGG (.ogg)
- And other formats supported by FFmpeg
Processing Details
Time Extraction
- Filename Parsing (Primary): Extracts numeric sequences from filename (e.g.,
note_20200101_040820.wav→ 2020-01-01 04:08:20) - File Modification Time (Fallback): Uses the file's
mtimewhen filename parsing fails
Coordinate Extraction
Time-Based Correlation Only: Audio files rely entirely on time correlation with GPS tracks. The system matches the audio timestamp with the nearest GPS track point within the configured time window.
Audio Processing
- Format Conversion: All audio is converted to MP3 format for web compatibility
- Transcription (Optional):
- Automatic speech-to-text with a local Whisper model — nothing is sent to a network service
- Creates timestamped segments
- Generates searchable text content
- Title Generation: An LLM served by ollama summarizes the transcript into a title
- Web Player: Creates interactive audio player with transcript display
Both steps are labelled as machine-generated in the journal and listed on the credits page's AI disclosure.
Configuration
features:
transcription:
enabled: !auto # Auto-detect: true when whisper is installed
llms:
enabled: true # Enable/disable LLM features
geo_correlation: # For coordinate correlation
max_time_diff: !duration 300 seconds # Max correlation window
llm_prompts:
generate_title: # Title generation settings
model: "granite3.3:8b" # The ollama model to use
options:
temperature: 0.2
The prompt text itself lives in strings.generate_title_prompt; see the
configuration reference. Per-recorder clock
offsets belong in a calibration.yaml, not in
features.geo_correlation.time_offset, which is not implemented.
Dependencies
- pydub: Audio format conversion
- FFmpeg: Backend for audio processing
- openai-whisper (
transcriptionextra): Local speech-to-text (optional) - Ollama: Title generation (optional)
Tips for Best Results
- Naming Convention: Use timestamp-based filenames for accurate time extraction
- Time Synchronization: Keep audio recording device time synced with GPS device
- Transcription Quality: Clear audio and supported language improve AI transcription accuracy
- GPS Logging: Maintain GPS tracks during audio recording for location data
Output
- Audio Player: Interactive web-based player in journal entries
- Transcription: Searchable text content with timestamps
- AI Titles: Automatically generated descriptive titles based on content
- Location Data: Coordinates from GPS correlation (when available)