Method and system for learning and using latent-space representations of audio signals for audio content-based retrieval
A method and system are provided for extracting features from digital audio signals which exhibit variations in pitch, timbre, decay, reverberation, and other psychoacoustic attributes and learning, from the extracted features, an artificial neural network model for generating contextual latent-space representations of digital audio signals. A method and system are also provided for learning an artificial neural network model for generating consistent latent-space representations of digital audio signals in which the generated latent-space representations are comparable for the purposes of determining psychoacoustic similarity between digital audio signals. A method and system are also provided for extracting features from digital audio signals and learning, from the extracted features, an artificial neural network model for generating latent-space representations of digital audio signals which take care of selecting salient attributes of the signals that represent psychoacoustic differences between the signals.
1. A method comprising:
reading a particular digital audio signal source of a set of digital audio signal sources associated with a particular sound content category;
generating a time-frequency representation based on the particular digital audio signal;
learning, using an artificial neural network, a set of numerical codes providing a latent-space representation of the time-frequency representation, wherein the set of numerical codes has a dimensionality that is less than a dimensionality of the time-frequency representation;
repeating the reading, generating, and learning for each of a plurality of other digital audio signal sources of the set of digital audio signal sources in the particular sound content category to train the artificial neural network;
obtaining a set of learned model parameters from the trained artificial neural network;
storing the set of learned model parameters for the particular sound content category in computer storage media;
generating a first time-frequency representation based on a first digital audio signal associated with the particular sound content category;
using the set of learned model parameters to compute a first set of numerical codes providing a latent-space representation of the first time-frequency representation;
generating a second time-frequency representation based on a second digital audio signal associated with the particular sound content category;
using the set of learned model parameters to compute a second set of numerical codes providing a latent-space representation of the second time-frequency representation; and
computing a distance between the first set of numerical codes and the second set of numerical codes.
2. The method of claim 1 , further comprising:
based on the distance, causing an indication to be presented in a computer graphical user interface that the first digital audio signal and the second digital audio signal encompass similar sounds.
3. The method of claim 1 , further comprising:
receiving the first digital audio signal over a data communications network from a computing device; and
wherein the first digital audio signal is captured by the computing device using a microphone of, or operatively coupled to, the computing device that records a human-audible performance.
4. The method of claim 1 , further comprising:
comparing the distance to a distance threshold; and
based on the distance being below the distance threshold, causing an indication to be presented in a computer graphical user interface that the first digital audio signal and the second digital audio signal encompass near-duplicate sounds.
5. The method of claim 4 , further comprising:
selecting the distance threshold based on the particular sound content category.
6. The method of claim 4 , further comprising:
computing the distance between the first set of numerical codes and the second set of numerical codes based on a cosine similarity between the first set of numerical codes and the second set of numerical codes.
7. The method of claim 1 , wherein the particular sound content category is selected from the group consisting of loops and one-shots.
8. The method of claim 1 , wherein the particular sound content category is selected from the group consisting of drum loops, drum one-shots, instrument loops, and instrument one-shots.
9. The method of claim 1 , wherein the artificial neural network comprises an input layer, one or more encoder intermediate layers, a bottleneck layer, one or more decoder intermediate layers, and an output layer; and wherein the method further comprises obtaining the set of numerical codes providing the latent-space representation of the time-frequency representation from the bottleneck layer.
10. The method of claim 9 , wherein the one or more encoder intermediate layers comprise one or more convolutional layers; and wherein the one or more decoder intermediate layers comprise one or more convolutional layers.
11. The method of claim 1 , wherein generating the time-frequency representation based on the particular digital audio signal is based on:
obtaining a pre-processed signal based on the particular digital audio signal;
generating a duration normalized signal based on the pre-processed signal;
selecting a time-shifted slice signal of the duration normalized signal, the time-shifted slice signal having a particular duration; and
generating the time-frequency representation based on the time-shifted slice signal.
12. The method of claim 11 , further comprising:
selecting the particular duration based on the particular sound content category.
13. The method of claim 1 , wherein generating the time-frequency representation based on the particular digital audio signal is based on:
generated a duration normalized signal based on the particular digital audio signal;
generating a pre-processed signal based on the duration normalized signal;
selecting a time-shifted slice signal of the pre-processed signal, the time-shifted slice signal having a particular duration; and
generating the time-frequency representation based on the time-shifted slice signal.
14. The method of claim 13 , further comprising:
selecting the particular duration based on the particular sound content category.
15. The method of claim 1 , further comprising:
generating a plurality of time-frequency representations based on the particular digital audio signal; and
learning, using the artificial neural network, a plurality of sets of numerical codes providing latent-space representations of the plurality of time-frequency representations.
16. The method of claim 15 , wherein the particular sound content category is a loop-type sound content category.
17. A computing system comprising:
one or more processors;
storage media; and
instructions stored in the storage media which, when executed by the computing system, cause the computing system to perform:
generating a plurality of time-frequency representations based on a plurality of digital audio signals associated with a particular sound content category;
training an artificial neural network to learn a plurality of sets of numerical codes providing a plurality of latent-space representations of the plurality of time-frequency representations;
obtaining a set of learned model parameters from the trained artificial neural network;
storing the set of learned model parameters for the particular sound content category in computer storage media;
generating a first time-frequency representation based on a first digital audio signal associated with the particular sound content category;
using the set of learned model parameters to compute a first set of numerical codes providing a latent-space representation of the first time-frequency representation;
generating a second time-frequency representation based on a second digital audio signal associated with the particular sound content category;
using the set of learned model parameters to compute a second set of numerical codes providing a latent-space representation of the second time-frequency representation; and
computing a distance between the first set of numerical codes and the second set of numerical codes.
18. The computing system of claim 17 , further comprising instructions which, when executed by the computing system, cause the computing system to perform:
based on the distance, causing an indication to be presented in a computer graphical user interface that the first digital audio signal and the second digital audio signal encompass similar sounds.
19. The computing system of claim 17 , further comprising instructions which, when executed by the computing system, cause the computing system to perform:
receiving the first digital audio signal over a data communications network from a computing device; and
wherein the first digital audio signal is captured by the computing device using a microphone of, or operatively coupled to, the computing device that records a human-audible performance.
20. The computing system of claim 17 , further comprising instructions which, when executed by the computing system, cause the computing system to perform:
comparing the distance to a distance threshold; and
based on the distance being below the distance threshold, causing an indication to be presented in a computer graphical user interface that the first digital audio signal and the second digital audio signal encompass near-duplicate sounds.