Systems and methods for segmenting and/or classifying an audio signal from transformed audio information
A system and method may be provided to segment and/or classify an audio signal from transformed audio information. Transformed audio information representing a sound may be obtained. The transformed audio information may specify magnitude of a coefficient related to energy amplitude as a function of frequency for the audio signal and time. Features associated with the audio signal may be obtained from the transformed audio information. Individual ones of the features may be associated with a feature score relative to a predetermined speaker model. An aggregate score may be obtained based on the feature scores according to a weighting scheme. The weighting scheme may be associated with a noise and/or SNR estimation. The aggregate score may be used for segmentation to identify portions of the audio signal containing speech of one or more different speakers. For classification, the aggregate score may be used to determine a likely speaker model to identify a source of the sound in the audio signal.
1. A system configured for segmenting and/or classifying an audio signal from transformed audio information, the system comprising:
one or more processors configured to execute computer program modules, the computer program modules comprising:
an audio information module configured to obtain transformed audio information representing a sound, wherein the transformed audio information specifies magnitude of a coefficient related to energy amplitude as a function of frequency for the audio signal and time;
an audio features module configured to obtain features associated with the audio signal from the transformed audio information, individual ones of the features being associated with a feature score relative to a predetermined speaker model; and
an aggregate score module configured to obtain an aggregate score based on the feature scores according to a weighting scheme, the weighting scheme being associated with a noise estimation, the aggregate score being used in conjunction with the predetermined speaker model to identify a source of the sound in the audio signal.
2. The system of claim 1 , further comprising a information reduction module configured to obtain reduced transformed audio information as a feature associated with the audio signal, wherein obtaining the reduced transformed audio information includes:
determining harmonic paths for individual harmonics of the sound based on fractional chirp rate and harmonic number;
determining an amplitude value for individual harmonics at individual time windows; and
constructing the reduced transformed audio information, wherein the reduced transformed audio information includes the amplitude values for individual harmonics as a function of frequency and time.
3. The system of claim 2 , wherein values within the reduced transformed audio information between the amplitude values are set to zero.
4. The system of claim 2 , wherein obtaining the reduced transformed audio information includes applying a filter bank to the reduced transformed audio information.
5. The system of claim 1 , further comprising a spectral slope module configured to obtain spectral slope information based on the transformed audio information as a feature associated with the audio signal, wherein obtaining the spectral slope information includes taking a derivative over frequency of the coefficient related to energy amplitude of a harmonic envelope in the transformed audio information.
6. The system of claim 5 , wherein obtaining the spectral slope information further includes applying a filter bank to the derivative.
7. The system of claim 1 , further comprising an SNR estimator module configured to obtain a noise and/or SNR estimation associated with the audio signal.
8. The system of claim 7 , wherein obtaining the noise estimation includes:
determining a harmonic energy of individual harmonics as a function of frequency and time based on the transformed audio information;
determining a total energy of the audio signal based on the transformed audio information;
determining a ratio between the harmonic energy and the total energy; and
estimating noise as portions of the transformed audio information where the ratio breaches a predetermined threshold.
9. The system of claim 7 , wherein obtaining the noise estimation further includes averaging the ratio over time.
10. The system of claim 1 , further comprising a training module configured to determine the weighing scheme, wherein the weighting scheme is determined based on one or more of a predetermined speaker model, a certain speaker, a set of speakers, or a certain audio capture device.
11. A computer-implemented method for segmenting and/or classifying an audio signal from transformed audio information, the method being implemented in a computer system that includes one or more physical processors, the method comprising:
obtaining, at the one or more physical processors, transformed audio information representing a sound, wherein the transformed audio information specifies magnitude of a coefficient related to energy amplitude as a function of frequency for the audio signal and time;
obtaining, at the one or more physical processors, features associated with the audio signal from the transformed audio information, individual ones of the features being associated with a feature score relative to a predetermined speaker model; and
obtaining, at the one or more physical processors, an aggregate score based on the feature scores according to a weighting scheme, the weighting scheme being associated with a noise estimation, the aggregate score being used in conjunction with the predetermined speaker model to identify a source of the sound in the audio signal.
12. The method of claim 11 , further comprising obtaining, at the one or more physical processors, reduced transformed audio information as a feature associated with the audio signal, wherein obtaining the reduced transformed audio information includes:
determining, on the one or more physical processors, harmonic paths for individual harmonics of the sound based on fractional chirp rate and harmonic number;
determining, on the one or more physical processors, an amplitude value for individual harmonics at individual time windows; and
constructing, on the one or more physical processors, the reduced transformed audio information, wherein the reduced transformed audio information includes the amplitude values for individual harmonics as a function of frequency and time.
13. The method of claim 12 , wherein values within the reduced transformed audio information between the amplitude values are set to zero.
14. The method of claim 12 , wherein obtaining the reduced transformed audio information includes applying a filter bank to the reduced transformed audio information.
15. The method of claim 11 , further comprising obtaining, at the one or more physical processors, spectral slope information based on the transformed audio information as a feature associated with the audio signal, wherein obtaining the spectral slope information includes taking a derivative over frequency of the coefficient related to energy amplitude of a harmonic envelope in the transformed audio information.
16. The method of claim 15 , wherein obtaining the spectral slope information further includes applying a filter bank to the derivative.
17. The method of claim 11 , further comprising obtaining, at the one or more physical processors, a noise and/or SNR estimation associated with the audio signal.
18. The method of claim 17 , wherein obtaining the noise estimation includes:
determining, on the one or more physical processors, a harmonic energy of individual harmonics as a function of frequency and time based on the transformed audio information;
determining, on the one or more physical processors, a total energy of the audio signal based on the transformed audio information;
determining, on the one or more physical processors, a ratio between the harmonic energy and the total energy; and
estimating, on the one or more physical processors, noise as portions of the transformed audio information where the ratio breaches a predetermined threshold.
19. The method of claim 17 , wherein obtaining the noise estimation further includes averaging the ratio over time.
20. The method of claim 11 , further comprising determining, on the one or more physical processors, the weighing scheme based on one or more of a predetermined speaker model, a certain speaker, a set of speakers, or a certain audio capture device.