How tearcake/audio-classifier works
This repository combines a browser audio-classification workflow with a separate PyTorch training path. Users select audio, the Next.js page trims/encodes it, and its upload route forwards it to FastAPI; inference preprocesses audio, runs the checkpoint-backed AudioCNN, and returns predictions and visualizations for the UI. Training builds the same model and saves its best checkpoint. The main inference, route, page, model, training, and feature-map relationships are supported by excerpts; other UI rendering capabilities are documented or visible in the page excerpt, while their implementation details were not sampled.
Main entry points: page.tsx Upload page · main.py Inference API · train.py Training pipeline · route.ts Upload proxy
Components 14
Browser application
- Upload page
audio-visualization/src/app/page.tsx - Audio preparation
audio-visualization/src/lib/audio.ts - Upload proxy
audio-visualization/src/app/api/upload/route.ts
Inference service
- Inference API
main.py - Audio preprocessing
main.py - Audio classifier
main.py
Model lifecycle
- AudioCNN
model.py - Best checkpoint
models/checkpoints/best_model.pth - Training pipeline
train.py
Result visualizations
- Feature-map view
audio-visualization/src/components/FeatureMap.tsx - Feature colors
audio-visualization/src/lib/colors.ts - Waveform view
audio-visualization/src/components/Waveform.tsx - Color scale
audio-visualization/src/components/ColorScale.tsx
Other components
- Audio uploader
Connections 14
- Audio uploader → Upload page: selects audio
README.md - Upload page → Audio preparation: prepares audio
audio-visualization/src/app/page.tsx - Upload page → Upload proxy: posts payload
audio-visualization/src/app/page.tsx - Upload proxy → Inference API: forwards request
audio-visualization/src/app/api/upload/route.ts - Inference API → Audio preprocessing: decodes and transforms
main.py - Inference API → Audio classifier: runs inference
main.py - Audio classifier → Best checkpoint: loads weights
main.py - Audio classifier → AudioCNN: invokes model
main.py - Training pipeline → AudioCNN: trains model
train.py - Training pipeline → Best checkpoint: saves best weights
train.py - Upload page → Feature-map view: renders maps
audio-visualization/src/app/page.tsx - Feature-map view → Feature colors: gets colors
audio-visualization/src/components/FeatureMap.tsx - Upload page → Waveform view: renders waveform
audio-visualization/src/app/page.tsx - Upload page → Color scale: renders scale
audio-visualization/src/app/page.tsx