Audio identification during performance
Methods and apparatus for audio identification during a performance are disclosed herein. An example method includes transforming a segment of audio into a log-frequency spectrogram based on a transform using a logarithmic frequency resolution. The log-frequency spectrogram is transformed into a binary image, each pixel of the binary image corresponding to a time frame and frequency channel pair. A matrix product of the binary image and a plurality of reference fingerprints is computed. Based on a number of frequency ranges represented in the binary image, the matrix product to form a similarity matrix is computed. Selecting an alignment of a line in the similarity matrix that intersects one or more bins in the similarity matrix with the largest calculated Hamming similarities. A reference fingerprint is selected based on the alignment.
1. A method for identifying audio, the method comprising:
transforming, by executing an instruction with a processor, a segment of audio into a log-frequency spectrogram based on a constant Q transform using a logarithmic frequency resolution;
transforming, by executing an instruction with the processor, the log-frequency spectrogram into a binary image, each pixel of the binary image corresponding to a time frame and frequency channel pair, each frequency channel representing a corresponding quarter tone frequency channel in a range from musical note C3 to musical note C8;
computing, by executing an instruction with the processor, a matrix product of the binary image and a plurality of reference fingerprints;
normalizing, by executing an instruction with the processor and based on a number of frequency ranges represented in the binary image, the matrix product to form a similarity matrix;
selecting, by executing an instruction with the processor, an alignment of a line in the similarity matrix that intersects one or more bins in the similarity matrix with the largest calculated Hamming similarities; and
selecting, by executing an instruction with the processor, a reference fingerprint based on the alignment.
2. The method of claim 1 , wherein the selecting of the alignment is performed without using a hash function.
3. The method of claim 1 , wherein the line has an angle within a range of forty to fifty degrees in the similarity matrix.
4. The method of claim 3 , wherein the angle of the line represents a tempo deviation between the segment of audio and the audio represented by the corresponding reference fingerprints.
5. The method of claim 1 , wherein the computing of the matrix product includes introducing one or more pitch shifts to the binary image.
6. The method of claim 5 , wherein the one or more pitch shifts correspond to key variations in a range of negative five semitones to positive five semitones.
7. The method of claim 1 , wherein the selecting of the best alignment is performed using a Hough Transform.
8. The method of claim 1 , wherein the computing of the matrix product is performed using a Hamming similarity.
9. The method of claim 1 , further including binarizing the similarity matrix based on a threshold value.
10. A tangible computer readable storage disk or storage device comprising instructions which, when executed, cause a machine to at least:
transform a segment of audio into a log-frequency spectrogram based on a constant Q transform using a logarithmic frequency resolution;
transform the log-frequency spectrogram into a binary image, each pixel of the binary image corresponding to a time frame and frequency channel pair, each frequency channel representing a corresponding quarter tone frequency channel in a range from musical note C3 to musical note C8;
compute a matrix product of the binary image and a plurality of reference fingerprints;
normalize, based on a number of frequency ranges represented in the binary image, the matrix product to form a similarity matrix;
select an alignment of a line in the similarity matrix that intersects one or more bins in the similarity matrix with the largest calculated Hamming similarities; and
select a reference fingerprint based on the alignment.
11. The tangible computer readable storage disk or storage device of claim 10 , wherein the selecting of the alignment is performed without using a hash function.
12. The tangible computer readable storage disk or storage device of claim 10 , wherein the line has an angle within a range of forty to fifty degrees in the similarity matrix.
13. The tangible computer readable storage disk or storage device of claim 12 , wherein the angle of the line represents a tempo deviation between the segment of audio and the audio represented by the corresponding reference fingerprints.
14. The tangible computer readable storage disk or storage device of claim 10 , wherein the instructions, when executed, cause the processor to compute the matrix product by introducing one or more pitch shifts to the binary image.
15. The tangible computer readable storage disk or storage device of claim 14 , wherein the one or more pitch shifts correspond to key variations in a range of negative five semitones to positive five semitones.
16. The tangible computer readable storage disk or storage device of claim 10 , wherein the instructions, when executed, cause the processor to select the best alignment using a Hough Transform.
17. The tangible computer readable storage disk or storage device of claim 10 , wherein the computing of the matrix product is performed using a Hamming similarity.
18. The tangible computer readable storage disk or storage device of claim 10 , wherein the instructions, when executed, cause the processor to binarize the similarity matrix based on a threshold value.