How aronjohn20/acoustic-localization works
This is a simulated acoustic-surveillance application: it generates multichannel room audio, detects sound events, localizes qualifying events, and can display and report session results. The principal runtime flow is owned by stream/pipeline.py, which consumes generated chunks, runs SED, conditionally computes GCC-PHAT features and runs DOA, then optionally sends results to the dashboard. Training and dataset-generation capabilities are included as separate branches. Some subsystem links rely on README or source-index references because those implementation files were not excerpted.
Main entry points: pipeline.py Inference orchestrator · generator.py Room stream generator · dashboard.py Live dashboard · train.py DOA model training
Components 18
Audio simulation
- System configuration
config.py - Room stream generator
stream/generator.py
Detection and localization
- Inference orchestrator
stream/pipeline.py - SED predictor
sed/api.py - CNN14 model
sed/model/main.py - DOA predictor
doa/api.py - SSL localization model
doa/model/main.py - Signal processing
dsp/main.py
Model training
- DOA data generation
doa/training_pipeline/data_gen/generate_data.py - DOA training settings
doa/training_pipeline/train_config.py - DOA model training
doa/training_pipeline/train.py - SED dataset
sed/training_pipeline/dataset.py - SED model training
sed/training_pipeline/train.py
Visualization and reporting
- Live dashboard
web_visualization/dashboard.py - Flask Socket.IO server
web_visualization/server.py - Browser visualization
web_visualization/static/js/app.js - Session report
web_visualization/report.py
Other components
- Operator
Connections 20
- Operator → Inference orchestrator: starts
README.md - Inference orchestrator → Room stream generator: starts generator
stream/pipeline.py - Room stream generator → Inference orchestrator: queues audio chunks
stream/pipeline.py - Room stream generator → System configuration: reads settings
stream/generator.py - Room stream generator → Signal processing: uses coordinates
stream/generator.py - Inference orchestrator → System configuration: reads thresholds
stream/pipeline.py - Inference orchestrator → SED predictor: runs detection
stream/pipeline.py - SED predictor → CNN14 model: runs model
sed/api.py - Inference orchestrator → Signal processing: computes GCC-PHAT
stream/pipeline.py - Inference orchestrator → DOA predictor: runs localization
stream/pipeline.py - DOA predictor → SSL localization model: runs model
doa/api.py - Inference orchestrator → Live dashboard: forwards results
stream/pipeline.py - Live dashboard → Flask Socket.IO server: posts updates
web_visualization/dashboard.py - Live dashboard → Session report: generates report
web_visualization/dashboard.py - Flask Socket.IO server → Browser visualization: broadcasts events
README.md - DOA data generation → DOA training settings: uses settings
doa/training_pipeline/data_gen/generate_data.py - DOA model training → DOA training settings: reads settings
doa/training_pipeline/train.py - DOA model training → SSL localization model: trains model
doa/training_pipeline/train.py - SED model training → SED dataset: loads dataset
sed/training_pipeline/train.py - SED model training → CNN14 model: trains model
sed/training_pipeline/train.py