User Tools

Site Tools


deepdreameffect

====== Differences ====== This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
deepdreameffect [2015/10/25 11:01]
dmr
deepdreameffect [2015/11/05 04:57] (current)
dmr
Line 4: Line 4:
 | **Affiliation** | International Audio Laboratories Erlangen | | **Affiliation** | International Audio Laboratories Erlangen |
 | **eMail** | [[christian.dittmar@audiolabs-erlangen.de]] | | **eMail** | [[christian.dittmar@audiolabs-erlangen.de]] |
 +| **code** | [[https://​github.com/​stefan-balke/​hamr2015-deepdreameffect]] |
  
-propose to use Google'​s DeepDream processing as an audio effect. Therefore, I export music magnitude spectrogram as RGB channels of PNG images and apply to '​Gradient Ascent'​ with pre-trained networks to these images. Since the networks were trained on natural images, this makes no sense musically. However, it gives interesting results:+===== What did do =====
  
 +I used Google'​s DeepDream processing as an audio effect. Therefore, I export music magnitude spectrogram as RGB channels of PNG images and apply so-called '​Gradient Ascent'​ with pre-trained networks to these images. Afterwards, I convert the resulting images to spectrograms again and resynthesize them using Griffin and Lim's method.
  
 +{{ :​overview.png?​nolink&​800 |}}
 +
 +Since the networks were trained on natural images, this makes no sense musically. However, it gives interesting results:
 +
 +===== Example 1: Piano =====
 +
 +Input signal {{ :shenua.wav |}}
 +Result using layer conv3 (MIT places network){{ :​output_shenhua_layer3.wav |}}
 +Result using layer pool5 (MIT places network){{ :​output_shenhua_layer5.wav |}}
 +
 +===== Example 2: Ethno =====
 +
 +Input signal {{ :olcay.wav |}}
 +Result using layer conv3 (MIT places network) {{ :​output_olcay_layer3.wav |}}
 +
 +===== Example 3: Breakbeat =====
 +
 +Input signal (Different drums encoded as RGB) {{ :​amenbrotherbreaknorm_mix.wav |}}
 +Result using layer conv3 (MIT places network) {{ :​output_amen_layer3.wav |}}
 +
 +===== Libraries Used =====
 +
 +    Anaconda Python Package
 +    Caffe Deep Deep Learning Framework
 +    Pre-Trained Networks
 +    iPython Notebook
 +    MATLAB
deepdreameffect.1445785304.txt.gz ยท Last modified: 2015/10/25 11:01 by dmr