tyiannak / color_your_music_mood

A realtime demo for generating colors based on musical moods

color_your_music_mood

This demo records sound and - when music is detected - it estimates the underlying mood (emotion) and based on that it generates a respective color. If available, it can even set your Yeelight Bulb color (again based on the detected musical mood)

Checkout this hackernoon article for more details on the implemented methodology and experimentation

Video Demo

How it works

The basic steps of this demo are the following:

1. Sound is recorded from the computer's microphone.
2. Every 5 seconds, the recorded audio segment is fed as input to two music emotion classifiers:
   • Valence, that expresses if the emotion is negative or positive
   • Energy, that expresses how strong an emotion is
3. Valence and Energy are represented in a 2D plane where Valence is the x-axis and Energy the y-axis.
4. A set of predefined points and colors is used do build the 2D colorspace. This set is defined as follow:
   • each pair of (point, color) is used to "paint" the 2D colorspace
   • all intermediate points are painted using a normalized linear combination of the predefined colors (based on the intermediate points' eucledian distances from the predefined colors)
5. The estimated Valence-Arousal pair for each 5-second music segment can be now represented with a particular color based on the 2D colored space.
6. The same color can be used to illuminate the smart Yeelight Bulbs, if available.

Some details on the used classifiers

• A music vs non-music classifier is applied as a first step
• Estimation of music emotional energy and valence is achieved through a simple SVM model
• The music classifiers have been trained on hand-crafted audio features, based on around 5000 audio training samples.
• All trained models are included in this repo.

Dependencies

• Connection to the Yeelight Bulbs is achieved using this library https://yeelight.readthedocs.io/en/latest/
• Audio analysis is performed using pyAudioAnalysis

color_your_music_mood's Issues

when I try to run your code , there is something wrong , do you know why?

---

AttributeError Traceback (most recent call last)
in
1 [clf_energy, mu_energy, std_energy, class_names_energy,mt_win_en, mt_step_en, st_win_en, st_step_en, _] = aT.load_model("energy")
2 fv_2 = (f_2[:, 0] - mu_energy) / std_energy
----> 3 [res_energy, p_en] = aT.classifier_wrapper(clf_energy, "svm_rbf", fv_2)
4 # win_class_energy = class_names_energy[int(res_energy)]

/data/miniconda2_python2.7/envs/py36_wy/lib/python3.6/site-packages/pyAudioAnalysis/audioTrainTest.py in classifier_wrapper(classifier, classifier_type, test_sample)
90 - model: regression model
91 - model_type: "svm" or "knn" (TODO)
---> 92 - test_sample: a feature vector (numpy array)
93 RETURNS:
94 - R: regression result (estimated value)

/data/miniconda2_python2.7/envs/py36_wy/lib/python3.6/site-packages/sklearn/svm/_base.py in predict(self, X)
583 """
584 check_is_fitted(self)
--> 585 if self.break_ties and self.decision_function_shape == 'ovo':
586 raise ValueError("break_ties must be False when "
587 "decision_function_shape is 'ovo'")

AttributeError: 'SVC' object has no attribute 'break_ties'

Move to last scikit-learn version

Hi,
First, thanks for your great work !!

Had created repo with some added features, see : https://github.com/zak-45/WLEDAudioSyncRTMood

Any chance that you upgrade to the last scikit-learn version ?

thanks