Systems, devices, and methods for assigning mood labels to musical compositions
Computer-based systems, devices, and methods for assigning mood labels to musical compositions are described. A mood classifier is trained based on mood-labeled musically-coherent segments of musical compositions and subsequently applied to automatically assign mood labels to musically-coherent segments of musical compositions. In both cases, the musically-coherent segments are generated using automated segmentation algorithms.
1. A computer-implemented method of labeling a musical composition with at least one mood label, the method comprising:
segmenting the musical composition into a sequence of non-uniform musically-coherent segments based on a sound of the musical composition, the sequence of musically-coherent segments including at least two musically-coherent segments selected from a group consisting of: an intro, a verse, a chorus, a solo, a middle8, a pre-chorus, a bridge, and an outro; and
employing a mood classifier to automatically assign at least one respective mood label to each respective musically-coherent segment.
2. The method of claim 1 wherein the mood classifier comprises processor-executable instructions and/or data stored in a non-transitory processor-readable storage medium that, when executed by at least one processor, cause the at least one processor to assign the at least one respective mood label to each respective musically-coherent segment of the musical composition, and wherein employing the mood classifier to automatically assign the at least one respective mood label to each respective musically-coherent segment includes executing, by the at least one processor, the processor-executable instructions and/or data stored in the non-transitory processor-readable storage medium communicatively coupled to the at least one processor to cause the at least one processor to assign the at least one respective mood label to each respective musically-coherent segment.
3. The method of claim 1 wherein the mood classifier comprises processor-executable instructions and/or data stored in a non-transitory processor-readable storage medium, wherein the processor-executable instructions and/or data are based at least in part on training data comprising a library of mood-labeled musically-coherent segments.
4. The method of claim 1 , further comprising:
labelling the musical composition with at least one additional mood label, the at least one additional mood label with which the musical composition is labelled corresponding to at least one of the mood labels assigned to at least one musically-coherent segment of the musical composition.
5. The method of claim 4 wherein labeling the musical composition with at least one additional mood label includes labeling the musical composition with all distinct mood labels assigned to all of the musically-coherent segments of the musical composition.
6. The method of claim 4 wherein labeling the musical composition with at least one additional mood label includes labeling the musical composition with a subset of all distinct mood labels assigned to all of the musically-coherent segments of the musical composition, the subset consisting of only X most frequently occurring distinct mood labels among all distinct mood labels assigned to all of the musically-coherent segments of the musical composition, where X is an integer and X≥1.
7. The method of claim 4 wherein labeling the musical composition with at least one additional mood label includes combining at least two of the respective mood labels assigned to respective ones of at least two of the musically-coherent segments into at least one subsuming mood label and labelling the musical composition with the at least one subsuming mood label.
8. The method of claim 1 wherein:
employing a mood classifier to automatically assign at least one respective mood label to each respective musically-coherent segment includes employing the mood classifier to automatically assign, to at least one musically-coherent segment, at least one mood label selected from a group consisting of: Energetic, Heroic, Happy, Relaxed, Calm, Dreamy, Awed, Romantic, Sensual, Sexual, Wistful, Sad, Melancholic, Depressed, Anxious, Fearful, Angry, Hateful, Restless, Whimsical, Amusing, Annoying, Tense, Beautiful, Relaxing, Serene, Energizing, Erotic, Desirous, Indignant, Defiant, Joyful, Cheerful, Depressing, Scary, Fearful, and Triumphant.
9. A mood classifier comprising;
processor-executable instructions and/or data stored in a non-transitory processor-readable storage medium, wherein when executed by at least one processor communicatively coupled to the non-transitory processor-readable storage medium, the mood classifier;
segments a musical composition into a sequence of non-uniform musically-coherent segments based on a sound of the musical composition, the sequence of musically-coherent segments including at least two musically-coherent segments selected from a group consisting of: an intro, a verse, a chorus, a solo, a middle8, a pre-chorus, a bridge, and an outro; and
automatically assigns at least one respective mood label to each respective musically-coherent segment wherein the processor-executable instructions and/or data are based at least in part on training data comprising a library of mood-labeled musically-coherent segments.
10. The mood classifier of claim 9 wherein the library of mood-labeled musically-coherent segments is generated from a library of musical compositions by a process that, for each musical composition in the library of musical compositions:
segments the musical composition into musically-coherent segments; and
assigns at least one mood label to each musically-coherent segment.
11. The mood classifier of claim 10 wherein the library of mood-labeled musically-coherent segments is generated from a library of musical compositions by a process that, for each musical composition in the library of musical compositions:
segments the musical composition into musically-coherent segments; and
assigns only one respective mood label to each musically-coherent segment.
12. The mood classifier of claim 9 wherein the at least one mood label is selected from a group consisting of: Energetic, Heroic, Happy, Relaxed, Calm, Dreamy, Awed, Romantic, Sensual, Sexual, Wistful, Sad, Melancholic, Depressed, Anxious, Fearful, Angry, Hateful, Restless, Whimsical, Amusing, Annoying, Tense, Beautiful, Relaxing, Serene, Energizing, Erotic, Desirous, Indignant, Defiant, Joyful, Cheerful, Depressing, Scary, Fearful, and Triumphant.
13. A computer-implemented method of training a mood classifier to automatically assign at least one mood label to a musical composition, the method comprising:
generating mood-labeled musically-coherent segments, wherein generating mood-labeled musically-coherent segments includes, for a set of musical compositions:
segmenting each musical composition into a sequence of non-uniform musically-coherent segments based on a sound of the musical composition, the sequence of musically-coherent segments including at least two musically-coherent segments selected from a group consisting of: an intro, a verse, a chorus, a solo, a middle8, a pre-chorus, a bridge, and an outro; and
assigning at least one mood label to each musically-coherent segment;
and
training the mood classifier based on the mood-labeled musically-coherent segments.
14. The method of claim 13 wherein assigning at least one mood label to each musically-coherent segment includes receiving, by at least one processor, an assignment of at least one respective mood label for each respective musically-coherent segment.
15. The method of claim 13 wherein assigning at least one mood label to each musically-coherent segment includes assigning only one respective mood label to each musically-coherent segment.
16. The method of claim 13 wherein:
assigning at least one mood label to each musically-coherent segment includes assigning, to each musically-coherent segment, at least one mood label selected from a group consisting of: Energetic, Heroic, Happy, Relaxed, Calm, Dreamy, Awed, Romantic, Sensual, Sexual, Wistful, Sad, Melancholic, Depressed, Anxious, Fearful, Angry, Hateful, Restless, Whimsical, Amusing, Annoying, Tense, Beautiful, Relaxing, Serene, Energizing, Erotic, Desirous, Indignant, Defiant, Joyful, Cheerful, Depressing, Scary, Fearful, and Triumphant.
17. The method of claim 13 wherein each mood-labeled musically-coherent segment comprises a respective set of digital data objects, and wherein training the mood classifier based on the mood-labeled musically-coherent segments includes extracting training features from parameters specified in the digital data objects.
18. The method of claim 13 wherein training the mood classifier based on the mood-labeled musically-coherent segments includes, for each mood-labeled musically-coherent segment, constructing a respective audio waveform and extracting training features from properties of the audio waveform.