Musical Training, but not Mere Exposure to Music, Drives the Emergence of Chroma Equivalence in Artificial Neural Networks
Neural networks need active music training to develop the ability to hear octave equivalence, challenging theories of innate perception.
A new study from researchers Lukas Grasse and Matthew S. Tata, published on arXiv, uses artificial neural networks (ANNs) to investigate a fundamental question in auditory neuroscience: is the perception of 'chroma equivalence'—hearing notes an octave apart as similar—innate or learned? The research employed representational similarity analysis on models like Wav2Vec 2.0 and Data2Vec, fine-tuning them under different conditions. The key finding was that models trained on a supervised music transcription task developed clear representations of chroma equivalence. In stark contrast, models that were merely exposed to music through self-supervised learning tasks, or trained on speech, developed 'pitch height' (the sense of notes being higher or lower) but not chroma equivalence. This provides strong computational evidence that the human ability to hear octave relationships is not an automatic byproduct of auditory exposure but emerges as a specialized cognitive computation to support the specific task of understanding music. The work highlights the power of using ANNs as 'synthetic brains' to test developmental theories, suggesting that active, goal-directed musical training is crucial for developing this sophisticated aspect of pitch perception, which has implications for music education and our understanding of how expertise shapes perception.
- Only ANNs trained on supervised music transcription developed chroma equivalence, not those trained via self-supervised learning on music.
- Models Wav2Vec 2.0 and Data2Vec all developed pitch height representations, showing a baseline auditory capability.
- The research supports the theory that chroma equivalence is a learned, task-specific computation, not an innate auditory percept.
Why It Matters
This challenges theories of innate music perception and underscores the importance of active training, not just exposure, for developing complex cognitive skills.