Dev Tools · 2h ago
CLAP Audio Encoder Tested on Environmental Sound Classification
A developer exported the CLAP audio encoder to ONNX and evaluated its embeddings on the ESC-50 dataset. The model produced 512-dimensional vectors from five-second sounds, achieving nearest-neighbor matches within the same class. The pipeline ran on Apple Silicon CPU, with code and data available for reproduction.
Meridian48 take
This is a practical benchmark for ONNX deployment of audio models, but the 1-NN accuracy on ESC-50 is not reported, leaving the real-world performance unclear.
Read the full reporting
Classifying Environmental Sounds with CLAP and ONNX Runtime: A 1-NN Evaluation on ESC-50 →
DEV Community
audio-embeddingsonnx-runtime