Evaluating Pretrained Music Embeddings for Cross-Performance Jazz Standard Recognition
2026-07-01 • Sound
SoundMachine Learning
AI summaryⓘ
The authors explore how to identify jazz standards—well-known jazz tunes—from different audio recordings that may sound quite different due to changes in speed, key, or instruments. They compare a model trained from scratch on spectrograms to models that use pre-trained music representations. They find that while the from-scratch model tends to memorize the training data, the pre-trained models perform better but can be influenced by who is playing. The authors also introduce a method to reduce this performer bias and suggest that recognizing jazz standards can help improve music understanding technologies.
jazz standardsmusic retrievalspectrogrampretrained modelscontrastive learningmusic embeddingsnearest-neighbor retrievaloverfittingfoundation modelsmusic understanding
Authors
Çağrı Eser
Abstract
Recognizing jazz standards from audio is a challenging form of tune-level music retrieval: different performances of the same standard may vary in tempo, key, arrangement, instrumentation, improvisational content, and even whether the head melody is present. We study this problem using a curated subset of the Jazz Trio Database designed for cross-performance standard recognition. We compare a from-scratch trained Harmonic CNN baseline against frozen pretrained music representations from recent music understanding foundation models, using both supervised probing and nearest-neighbor retrieval. Our results suggest that from-scratch spectrogram models overfit strongly to training performances, while pretrained embeddings provide better top-$k$ results but are sensitive to performer identity, which can be partially reduced with a lightweight contrastive projection. Our findings motivate jazz standard recognition as a useful stress test for music representation models and as a step toward retrieval-based standard identification. Project page: https://github.com/cagries/tipofmyear.