CUIDADO Review Meeting 18'3'02: BGU Contribution

1
CUIDADO Review Meeting 18.3.02 BGU Contribution

• WP2.1.7.4 Dynamic Sound Descriptions Matching
  Between Low and High Level Features
• WP2.1.3.2 ICA and ISA signal analysis Study
  Report, Application for Classification
• WP2.1.09. Preliminary Audio Descriptors HOS for
  instrumental and sound FX matching

2
Complete Melody Likelihoodfor Query By Melody
Dynamic Sound Descriptions Matching Between
Sound and Higher Level Features CUIDADO WP2.1.7

• Shai Shalev and Shlomo Dubnov

3
Goals

• Matching between a high feature (score) and a
  low feature (sound).
• Modeling dynamic features (Tempo).

Solution Method

• Using Bayesian Network.
• Define latent segmentation variables.
• Temporal Markov assumptions.
• Likelihood measure for pitch.

P(Aligned Score Tempo, Score)
P(Sound Aligned Score, Instrument)
4
Results
5
Results

• Average Precision 0.90
• (av. num. of hits in the rank list)
• Coverage 1.82
• (av. Index in the ranked list to find all
  hits)
• One Error 0.18
• (Prob. to fail in the first position)

6
ICA methods and its applications in sound
analysis WP2.1.3.2 ICA and ISA signal analysis
Adiel Ben-Shalom and Shlomo Dubnov
7
Different approaches for Feature Extraction
Bell Sejnowski
Casey Representation Indep.
Coeff. Indep. Coeff. Transformation
None Magnitude
Spectra Order No data
reduction Data reduction (PCA) Result
Local sound
Frequency domain
waveforms Features
8
ICA basis functions
9
PCA basis functions
10
Sound Classification using ICA

• Features vector Mag. Spectrum ICA coefficients
• Discriminate between different classes of sounds
  using K-NN algorithm
• Successfully discriminate between male and female
• Test other
• A common database is needed to continue the
  classification test

11
HOS for instrumental and sound FX matching
Preliminary Audio Descriptors CUIDADO WP2.1.09

• Shlomo Dubnov

12
Kurtosis Profile for Musical Instruments
13
HOS Matching for Sound Effects
14
Classification Results

• Tested on a set of 11 sound files
• For 6 sounds the matching was 100.
• For other sounds, the correct matching was 95,
  85, 80, 70 and 60
• Efficient Implementation with Matched Filter