INSTRUMENT TECHNIQUE CLASSIFIER

2025

Built a convolutional neural network to classify instrument playing techniques and articulation patterns — such as staccato, legato, vibrato, and pluck vs. bow — across multiple instrument families from spectrogram features derived from raw audio.

• Designed and trained a CNN architecture to detect technique and articulation classes across guitar, piano, strings, and wind instruments
• Preprocessed audio into mel-spectrogram features and augmented datasets to improve generalization across timbres and recording conditions
• Evaluated model performance with per-class accuracy metrics and confusion analysis to identify weak articulation boundaries
• Built an inference pipeline for real-time technique prediction from live or recorded audio input