Presentation is loading. Please wait.

Presentation is loading. Please wait.

Creating Sound Texture through Wavelet Tree Learning and Modeling

Similar presentations


Presentation on theme: "Creating Sound Texture through Wavelet Tree Learning and Modeling"— Presentation transcript:

1 Creating Sound Texture through Wavelet Tree Learning and Modeling
By Guy Shterlich Asaf Kacen Supervisor Dr. Shlomo Dubnov

2 Motivation and Background
In simple words: create a synthesis sound texture from an original sample. What for? automatic generation of sound effects. creative musical and sonic manipulations. virtual reality sonification. Why? Simple repetition is not efficient and sounds artificial. Automation. Low storage.

3 Process steps Input: sound signals that are approximately stationary: rain, waterfall, traffic noises, people babble, machine noises, etc., Discrete Wavelet Transform: construct a tree, representing a hierarchical wavelet transform of the signal. Synthesis algorithm: new random trees are generated by learning and sampling the conditional probabilities of the paths in the original tree. Inverse DWT: Transformation of these random trees back into signals results in new sound textures that closely resemble the sonic impression of the original sound source but without exactly repeating it.

4 Discrete Wavelet Transform
Original signal Scaling filter Downsample x2 Wavelet filter Downsample x2 Scaling filter Downsample x2 Wavelet filter Downsample x2 Scaling Down x2 Wavelet Down x2 scaling (lev 0) detail (lev 1) detail coefficients (level 2) detail coefficients (level 3)

5 Tree Synthesis algorithm
Original tree new tree ?

6 Tree Synthesis algorithm
Original tree new tree ?

7 Candidate set Original tree new tree ?

8 Candidate set Original tree new tree ?

9 For i=1 to length(candidate set)
Original tree new tree ?

10 Anc Candidate Original tree new tree ?

11 Anc Candidate Original tree new tree ?

12 Anc Candidate Original tree new tree ?

13 Pre Candidate Original tree new tree ?

14 Pre Candidate Original tree new tree ?

15 Pre Candidate Original tree new tree ?

16 The winners are… Original tree new tree ?

17 Randomly choose Original tree new tree ?

18 Inherit candidates Original tree new tree

19 Inherit candidates Original tree new tree

20 CONTROL! Threshold K Originality Creativity Noisiness threshold

21 Demonstration skip

22

23

24

25

26 Conclusions & Future work
Algorithm limitation: Sound texture types dependency. Require delicate and fine control (automation!). Noise and distortions Vs. creativity . Artifacts in the reconstructed signal caused by even small changes in the wavelet domain. Future work: Enhance predecessor candidate algorithm . Larger Input. Research and uses of other wavelet types. GUI improvements.

27 Questions ?


Download ppt "Creating Sound Texture through Wavelet Tree Learning and Modeling"

Similar presentations


Ads by Google