TypeScript Audio Encoder Blueprint

This guide shows how to replicate the Python librosa_encoder.py functionality using TypeScript and ffmpeg.

Goals

• Trim or loop audio to exactly 12 seconds
• Resample to 44.1kHz mono
• Normalize loudness via ffmpeg's loudnorm filter
• Export as WAV
• Generate a simple mel-spectrogram array for frontend visuals

Implementation Outline

1. Preprocess with FFmpeg

   • Convert to mono, resample, trim and normalize:

   ffmpeg -y -i input.wav -ac 1 -ar 44100 -t 12 -filter:a loudnorm=I=-16:LRA=11:TP=-1 output.wav

2. Parse WAV

   • Read the PCM samples from the processed file.

3. Compute Spectral Features

   • Use a naive DFT to calculate magnitudes for each frame.
   • Average the magnitudes into nMels buckets to approximate a mel‑spectrogram.

Example Script

backend/audio_encoder.ts implements this flow. Run it with Bun:

bun backend/audio_encoder.ts input.wav processed.wav metadata.json

It writes the normalized clip and a JSON metadata file containing duration and an averaged mel-spectrogram array.