Speaker identification using spatial information

1. A method of speaker identification for audio content, the audio content being of a format based on multiple channels, the method comprising:

extracting, from a first audio clip in the format, a plurality of spatial acoustic features across the multiple channels and location information, the first audio clip including a plurality of frames for each of a plurality of channels, the first audio clip including audio content corresponding to voices from a speaker, the spatial acoustic features including acoustic characteristics of the voices from the speaker;

constructing a first model for the speaker based on the spatial acoustic features and the location information, the first model indicating a characteristic of the voices from the speaker; and

identifying whether the audio content contains voices from the speaker based on the first model, wherein the spatial acoustic features include an intra-channel shifted delta cepstrum (SDC) feature and an inter-channel SDC feature, and wherein extracting the spatial acoustic features from the first audio clip comprises:

for each of the multiple channels, extracting a cepstrum coefficient for each frame of the first audio clip in a frequency domain;

determining an intra-channel SDC feature for each of the multiple channels based on difference between the cepstrum coefficients for the channel over a predetermined number of frames; and

determining an inter-channel SDC feature for each two of the multiple channels based on difference between the cepstrum coefficients for the two channels.

2. The method according to claim 1 , wherein the location information includes at least one of a direction of arrival or a distance of the speaker relative to an audio recording device, and

wherein constructing a first model for the speaker based on the spatial acoustic features and the location information comprises:

determining a generalized linear discriminant sequence (GLDS) kernel function based on the spatial acoustic features; and

constructing the first model based on the location information and the GLDS kernel function.

3. The method according to claim 2 , wherein determining the GLDS kernel function based on the spatial acoustic features comprises:

combining the spatial acoustic features;

performing dimensional mapping on the combined spatial acoustic feature; and

determining the GLDS kernel function based on the mapped spatial acoustic feature.

4. The method according to claim 2 , wherein determining the GLDS kernel function based on the spatial acoustic features comprises:

performing dimensional mapping on each of the spatial acoustic features;

combining the mapped spatial acoustic features; and

determining the GLDS kernel function based on the combined spatial acoustic feature.

5. The method according to claim 1 , wherein identifying whether the audio content contains voices from the speaker based on the first model comprises:

extracting a plurality of spatial acoustic features across the multiple channels from the audio content;

determining a closeness score between the spatial acoustic features of the audio content and the first model; and

identifying whether the audio content contains voices from the speaker based on the closeness score.

6. The method according to claim 1 , wherein the speaker is a first speaker, and the method further comprising:

determining whether voices from a second speaker contain in the first audio clip based on the location information;

constructing a second model for the second speaker based on the spatial acoustic features and the location information, the second model indicating a characteristic of the voices from the second speaker; and

identifying whether the audio content contains voices from the second speaker based on the second model.

7. The method according to claim 1 , further comprising:

updating the first model based on a second audio clip of the format.

8. The method according to claim 7 , wherein updating the first model based on the second audio clip comprises:

extracting a plurality of spatial acoustic features across the multiple channels from the second audio clip;

determining a weight for the first model based on the spatial acoustic features of the second audio clip; and

combining the spatial acoustic features of the second audio clip to the first model based on the determined weight.

9. The method according to claim 8 , wherein determining a weight for the first model based on the spatial acoustic features of the second audio clip comprises:

determining a closeness score between the spatial acoustic features of the second audio clip and the first model; and

determining the weight for the first model based on the closeness score.

10. The method according to claim 8 , wherein determining a weight for the first model based on the spatial acoustic features of the second audio clip comprises:

constructing a third model for the speaker based on the spatial information, the third model being a Gaussian mixture model;

iteratively determining, in an estimation-maximization (EM) process, a probability of the second audio clip containing voices from the speaker based on the third model; and

determining the weight based on the probability.

11. The method according to claim 10 , wherein determining the weight based on the probability comprises:

determining a closeness score between the spatial acoustic features of the second audio clip and each first model; and

determining the weight based on the probability and the closeness score.

12. The method according to claim 10 , further comprising:

updating the third model in each iteration of the EM process based on the probability determined in the iteration.

13. The method according to claim 7 , wherein updating the first model based on the second audio clip further comprises:

determining a closeness score between the spatial acoustic features of the second audio clip and each of the first model and a second model constructed from the first audio clip;

merging the first model with the second model if the closeness scores of the first model and of the second model are higher than a first threshold; and

constructing a new model based on the spatial acoustic feature of the second audio clip if the closeness scores of the first model and the second model are both lower than a second threshold.

14. A computer program product for speaker identification for audio content, the computer program product being tangibly stored on a non-transitory computer-readable medium and comprising machine executable instructions which, when executed, cause the machine to perform steps of the method according to claim 1 .

15. A non-transitory computer readable medium having software stored thereon, the software including instructions for controlling one or more devices to perform a method of speaker identification for audio content, the audio content being of a format based on multiple channels, the method comprising:

extracting, from a first audio clip in the format, a plurality of spatial acoustic features across the multiple channels and location information, the first audio clip including a plurality of frames for each of a plurality of channels, the first audio clip including audio content corresponding to voices from a speaker, the spatial acoustic features including acoustic characteristics of the voices from the speaker;

constructing a first model for the speaker based on the spatial acoustic features and the location information, the first model indicating a characteristic of the voices from the speaker; and

identifying whether the audio content contains voices from the speaker based on the first model, wherein the location information includes at least one of a direction of arrival or a distance of the speaker relative to an audio recording device and wherein constructing a first model for the speaker based on the spatial acoustic features and the location information comprises:

determining a generalized linear discriminant sequence (GLDS) kernel function based on the spatial acoustic features; and

constructing the first model based on the location information and the GLDS kernel function.

16. The non-transitory computer readable medium of claim 15 , wherein determining the GLDS kernel function based on the spatial acoustic features comprises:

combining the spatial acoustic features;

performing dimensional mapping on the combined spatial acoustic feature; and

determining the GLDS kernel function based on the mapped spatial acoustic feature.

17. The non-transitory computer readable medium of claim 15 , wherein determining the GLDS kernel function based on the spatial acoustic features comprises:

performing dimensional mapping on each of the spatial acoustic features;

combining the mapped spatial acoustic features; and

determining the GLDS kernel function based on the combined spatial acoustic feature.