Method and system for extracting audio features from an encoded bitstream for audio classification
View Patent ↗A method and system for extracting audio features from an encoded bitstream for audio classification. The method comprises partially decoding the encoded bitstream; obtaining uniform window block size spectral coefficients of the encoded bitstream; and extracting audio features based on the uniform window block spectral coefficients.
1. A method for extracting audio features from an encoded bitstream for audio classification, the method comprising:
partially decoding the encoded bitstream using a decoder to obtain spectral coefficients of the encoded bitstream for audio classification;
obtaining uniform window block size spectral coefficients of the encoded bitstream by processing said spectral coefficients using a processor;
extracting audio features based on the uniform window block spectral coefficients for audio classification; and
classifying the encoded bitstream using the audio features,
wherein the obtaining the uniform window block size spectral coefficients comprises representing a short window block in the encoded bitstream by a mean value of absolute spectral coefficients over all the short windows in said short window block and then expanding dimensions of short window block size spectral coefficients of said short window block of the encoded bitstream to dimensions of long window block size spectral coefficients of said short window block of the encoded bitstream so as to obtain uniform window block size spectral coefficients.
2. A method as claimed in claim 1 , wherein the extracting of the audio features comprises:
subjecting the uniform window block spectral coefficients to a filter bank; and
applying a discrete cosine transform to extract quasi-Mel Frequency Cepstral Coefficients (quasi-MFCC).
3. A method as claimed in claim 1 , wherein the obtaining the uniform window block size spectral coefficients comprises utilising interpolation techniques to expand dimensions of short window block size spectral coefficients to dimensions of long window block size spectral coefficients.
4. A method for extracting audio features from an encoded bitstream for audio classification, the method comprising:
partially decoding the encoded bitstream using a decoder to obtain spectral coefficients of the encoded bitstream for audio classification;
obtaining uniform window block size spectral coefficients of the encoded bitstream by processing said spectral coefficients using a processor;
extracting audio features based on the uniform window block spectral coefficients for audio classification,
classifying the encoded bitstream using the audio features,
wherein the extracting of the audio features comprises:
calculating a global spectrum centroid based on the uniform window block spectral coefficients; and
deriving a Successive Lower Centroid (SLC) feature by reducing a spectrum range of the global spectrum centroid to a range based on a predefined convergence threshold.
5. A method as claimed in claim 4 , wherein a Centroid-Centroid Deviation feature is calculated by taking a difference between the global spectrum centroid and the Successive Lower Centroid.
6. A method as claimed in claim 5 , wherein the obtaining the uniform window block size spectral coefficients comprises representing a short window block in the encoded bitstream by a mean value of absolute spectral coefficients over all the short windows in said short window block.
7. A method as claimed in claim 6 , wherein the obtaining the uniform window block size spectral coefficients comprises utilising interpolation techniques to expand dimensions of short window block size spectral coefficients to dimensions of long window block size spectral coefficients.
8. A system for extracting audio features from an encoded bitstream for audio classification, the system comprising:
a decoder for partially decoding the encoded bitstream using a decoder to obtain spectral coefficients of the encoded bitstream for audio classification;
a processor for obtaining uniform window block size spectral coefficients of the encoded bitstream by representing a short window block in the encoded bitstream by a mean value of absolute spectral coefficients over all the short windows in said short window block and then expanding dimensions of short window block size spectral coefficients of said short window block of the encoded bitstream to dimensions of long window block size spectral coefficients of said short window block of the encoded bitstream so as to obtain uniform window block size spectral coefficients;
an audio feature extractor for extracting audio features based on the uniform window block spectral coefficients for audio classification; and
a classifier for classifying the encoded bitstream using the audio features.
9. A system as claimed in claim 8 , wherein the audio feature extractor comprises a filter bank for filtering the uniform window block spectral coefficients; and the processor applies discrete cosine transform to extract quasi-Mel Frequency Cepstral Coefficients (quasi-MFCC).
10. A system as claimed in claim 8 , wherein the processor utilises interpolation techniques to expand dimensions of short window block size spectral coefficients to dimensions of long window block size spectral coefficients.
11. A system for extracting audio features from an encoded bitstream for audio classification, the system comprising:
a decoder for partially decoding the encoded bitstream using a decoder to obtain spectral coefficients of the encoded bitstream for audio classification;
a processor for obtaining uniform window block size spectral coefficients of the encoded bitstream by processing said spectral coefficients;
an audio feature extractor for extracting audio features based on the uniform window block spectral coefficients for audio classification; and
a classifier for classifying the encoded bitstream using the audio features,
wherein the audio feature extractor calculates a global spectrum centroid based on the uniform window block spectral coefficients and derives a Successive Lower Centroid (SLC) feature by reducing a spectrum range of the global spectrum centroid to a range based on a predefined convergence threshold.
12. A system as claimed in claim 11 , wherein the audio feature extractor further calculates a Centroid-Centroid Deviation feature by taking a difference between the global spectrum centroid and the Successive Lower Centroid.
13. A system as claimed in claim 12 , wherein the processor represents a short window block in the encoded bitstream is represented by a mean value of absolute spectral coefficients over all short windows in said short window block.
14. A system as claimed in claim 13 , wherein the processor utilises interpolation techniques to expand dimensions of short window block size spectral coefficients to dimensions of long window block size spectral coefficients.