Elephant Paired-Rumble Pipeline

This main branch now uses the paired-rumble U-Net ML stack from hamiz-machine-learning as the only machine-learning implementation in the repo. The old scripts/elephant_pipeline and src/elephant_pipeline codepaths were removed.

ML Surface

• Model package: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\elephant_rumble_ml
• Config: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\configs\paired_rumble.yaml
• Training-data builder: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\build_training_data.py
• Manifest builder: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\prepare_paired_manifests.py
• Training entrypoint: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\train_paired_rumble_model.py
• Single-file inference: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\run_paired_rumble_model.py
• Batch inference: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\run_paired_rumble_batch.py
• Viewer catalog builder: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\build_viewer_catalog.py
• Convenience CLI: C:\Users\shrek\Documents\Elephant-Audio-Cleaning-and-Classification-Pipeline\scripts\run_all.py

Common Commands

Build paired training data from the official audio and labels:

python scripts/run_all.py build-training-data --audio-dir "Audio Files (04-10-2026)" --labels-csv "Documentation/Audio Files Master (04_10_2026) - 20260324_rumbles_in_noise_for_hackathon.csv"

Build paired manifests:

python scripts/run_all.py prepare-manifests --config configs/paired_rumble.yaml

Train the paired-rumble U-Net:

python scripts/run_all.py train --config configs/paired_rumble.yaml

Run batch inference over the official WAV set:

python scripts/run_all.py batch-infer --config configs/paired_rumble.yaml --audio-dir "Audio Files (04-10-2026)" --labels-csv "Documentation/Audio Files Master (04_10_2026) - 20260324_rumbles_in_noise_for_hackathon.csv"

Build the viewer catalog:

python scripts/run_all.py build-viewer

Start the UI:

cd webui
npm install
npm run dev

Viewer Surface

The webui app now reads paired-rumble outputs from artifacts/paired_rumble/outputs and shows:

• original audio
• cleaned paired-rumble output
• live waveform and spectrogram views
• saved model-generated before/after spectrogram images
• interval-only composite scores on the labeled elephant window
• paired-rumble training summary and checkpoint metadata

Generated runtime outputs stay under artifacts/.