Show HN: Classify mechanical faults using Contrastive Language-Audio Pretraining
Category: ai-ml
Tags: audio-ml, car-diagnosis, clap
Score: 7.3/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 6)
Cardiag is a proof-of-concept audio-ML pipeline that uses Contrastive Language-Audio Pretraining (CLAP) to diagnose car faults from sound clips. It is interesting because it combines scraping, audio cleaning, and honest model calibration to create a triage tool that knows when it is uncertain, achieving meaningful results despite the difficulty of the problem. The reusable cleaning and training recipe is a notable contribution for similar audio tasks.
Target audience: backend devs, data engineers
Repository: https://github.com/adam-s/car-diagnosis · Python · MIT
View on Hacker News