10420CS 573100 音樂資訊檢索 Music Information Retrieval
HW1 Discussion Yi-Hsuan Yang Ph.D. http://www.citi.sinica.edu.tw/pages/yang/
[email protected]
Music & Audio Computing Lab, Research Center for IT Innovation, Academia Sinica
HW1: Instrument Recognition • Required: disambiguating four instruments (clip-level) • Bonus: detecting the occurrence of the instruments (frame-level)
Baseline Approach: MFCC + RBF-SVM spectrogram dim: 1025 x 1292
mel spectrogram dim: 128 x 1292
Baseline Approach: Result
Your Approaches: Result
4 1
85.5% 83.0% 82.0% 81.0% 80.0%
Improvement: Characterizing Vibrato? • Q: How to characterize attack, vibrato etc?
Improvement: Note-based processing • Ans: need to know pitch & onset (W7, W8, W12) onset
pitch
offset
Improvement: Note-based processing
• Electric guitar playing technique detection in real-world recording based on F0 sequence pattern recognition, ISMIR 2015 • Analysis of expressive musical terms in violin using score-informed and expression based audio features, ISMIR 2015
Improvement: Why Pooling? • Q: Information lost due to pooling
mean, std, etc …
Improvement: Dictionary-based Approach • Ans: use codebooks (W9, W10)
codebook
Improvement: Dictionary-based Approach • Ans: use codebooks => also good for tracking codebook codebook codebook
Improvement: Dictionary-based Approach
dictionarybased approach Sparse cepstral and phase codes for guitar playing technique classification, ISMIR 2014
Bonus Task: Result • Evaluate in terms of F-score students'
baseline
0.4257 0.4010 0.3544 0.3269 0.3066 0.2959 0.2959 0.2862 0.1893 0.0901
0.2756
Bonus Task: Groundtruth
Bonus Task: Student’s Result (F-score 0.42)