Near-Optimal Discretization of the Brachistochrone Problem - CCRMA
CCRMA MIR2014 Wavelets
Transcript of CCRMA MIR2014 Wavelets
![Page 1: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/1.jpg)
Leigh M. Smith Humtap Inc.
CCRMA MIR Workshop 2014 Wavelets and multiresolution
representations
![Page 2: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/2.jpg)
Segmentation (Frames, Onsets,
Beats, Bars, Chord Changes, etc)
Feature Extraction
(Time-based, spectral energy,
MFCC, etc)
Analysis / Decision Making
(Classification, Clustering, etc)
Basic system overview
![Page 3: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/3.jpg)
3
Short Term Fourier Transform
Time
Frequency Time-Frequency Grid of the STFT
Minimum TimeResolution
MinimumDiscriminableFrequency
![Page 4: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/4.jpg)
Continuous wavelet transform (CWT) decomposes (invertibly) a signal onto scaled and translated instances of a finite time “mother function” or “basis”.
-30 -15 15 30 a = 1
-1
1Real
-30 -15 15 30
-1
1Imaginary
-30 -15 15 30 a = 2
-1
1
-30 -15 15 30
-1
1
Wavelet time-frequency analysis
Ws(b, a) =1pa
Z 1
�1s(⌧) · g(
⌧ � b
a) d⌧ , a > 0 (1)
g(t) = e�t2/2 · ei!0t (2)
1
4
![Page 5: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/5.jpg)
Example: Sinusoidal Signal
5
![Page 6: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/6.jpg)
Example: Sinusoidal Signal
6
3π/2
Phase Mapping:
π
π/2
0
![Page 7: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/7.jpg)
Example: Simple RhythmScaleogram and Phaseogram of an isochronous pulse rhythmic signal:
-30 -15 15 30 45Real
-1
1Beat n Beat n+1
-30 -15 15 30 45Imag
-1
1Beat n Beat n+1
7
![Page 8: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/8.jpg)
Implementation• Implemented as a set of complex value
bandpass filters in Fourier domain. • Scaling produces a “zooming” time window
for each frequency “scale”. • Creates simultaneous time and frequency
localisation close to the Heisenberg inequality.
8
![Page 9: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/9.jpg)
Wavelet Time-Frequency Resolution from Dilation (“Zooming'')
9
2000 4000 6000 8000Time in Samples2048
1024
512
256
128
64
32
16
8
4
2
Scale as IOI Range in Samples
![Page 10: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/10.jpg)
Wavelets for Rhythm (Smith & Honing 2008)
• The CWT enables representation of temporal structure in terms of time varying rhythmic frequencies.
– Produces magnitude and phase measures which reveal time-frequency ridges indicating the frequencies present in the input rhythm signal (collectively a skeleton, Tchamitchian & Torrésani ’92).
10
![Page 11: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/11.jpg)
Musical Example■ The rhythm of “Greensleeves”...
11
![Page 12: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/12.jpg)
Greensleeves
12
![Page 13: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/13.jpg)
Memory Based TactusWavelet rhythm analysis is also applicable to continuous onset salience traces from auditory models (Coath, et. al 2009).
13
![Page 14: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/14.jpg)
Memory Based Tactus
• Uses lossy windowed integrator to amass tactus likelihood.
• Suppress all but the magnitude coefficients of the extracted tactus ridge.
• Invert the extracted tactus ridge and original phase plane back to the time domain. Creates a single beat oscillation.
• Nominating a starting beat and noting its phase, all other foot-taps are generated for the same phase value.
14
![Page 15: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/15.jpg)
Reconstructed Phase
15
![Page 16: CCRMA MIR2014 Wavelets](https://reader031.fdocuments.us/reader031/viewer/2022032902/623fcdff0ef4c74e5e4a75be/html5/thumbnails/16.jpg)
• Singing examples of Dutch folk songs from the "Onder de Groene Linde" collection (Meertens Institute).
• Uses continuous wavelet transform of rhythmic signals (Smith 1996, Smith & Honing 2008) to derive tactus:
• Example 1: • Example 2: ...Original + Accompaniment.
Example: Foot-tapping to singing
+ Accompaniment.Original...
16