Methods and apparatuses for segmenting an audio-visual recording using image similarity searching and audio speaker recognition
First Claim
1. A method of segmenting an audio-video recording, comprising the steps of:
- identifying one or more video frame intervals having similarity to a predetermined video image class;
extracting one or more audio intervals corresponding to the one or more video frame intervals;
applying an acoustic clustering method on the one or more audio intervals to produce one or more audio clusters; and
wherein the step of identifying one or more video frame intervals comprises decimating a video portion of the audio-visual recording in time and space to produce decimated frames; and
for each decimated frame, transforming the decimated frame to produce a transform matrix;
extracting a feature vector from the transform matrix; and
determining similarity of the frame using the feature vector and a video image class statistical model.
7 Assignments
0 Petitions
Accused Products
Abstract
Methods for segmenting audio-video recording of meetings containing slide presentations by one or more speakers are described. These segments serve as indexes into the recorded meeting. If an agenda is provided for the meeting, these segments can be labeled using information from the agenda. The system automatically detects intervals of video that correspond to presentation slides. Under the assumption that only one person is speaking during an interval when slides are displayed in the video, possible speaker intervals are extracted from the audio soundtrack by finding these regions. Since the same speaker may talk across multiple slide intervals, the acoustic data from these intervals is clustered to yield an estimate of the number of distinct speakers and their order. Clustering the audio data from these intervals yields an estimate of the number of different speakers and their order. Merged clustered audio intervals corresponding to a single speaker are then used as training data for a speaker segmentation system. Using speaker identification techniques, the full video is then segmented into individual presentations based on the extent of each presenter'"'"'s speech. The speaker identification system optionally includes the construction of a hidden Markov model trained on the audio data from each slide interval. A Viterbi assignment then segments the audio according to speaker.
-
Citations
19 Claims
-
1. A method of segmenting an audio-video recording, comprising the steps of:
-
identifying one or more video frame intervals having similarity to a predetermined video image class;
extracting one or more audio intervals corresponding to the one or more video frame intervals;
applying an acoustic clustering method on the one or more audio intervals to produce one or more audio clusters; and
wherein the step of identifying one or more video frame intervals comprises decimating a video portion of the audio-visual recording in time and space to produce decimated frames; and
for each decimated frame, transforming the decimated frame to produce a transform matrix;
extracting a feature vector from the transform matrix; and
determining similarity of the frame using the feature vector and a video image class statistical model. - View Dependent Claims (2, 3)
subtracting a video image class mean vector from the feature vector to determine a difference vector; and
comparing a magnitude of the difference vector to a threshold.
-
-
3. A method as in claim 2, wherein the step of comparing a magnitude of the difference vector to a threshold comprises the step of:
comparing the magnitude of the difference vector to a predetermined multiple of a standard deviation associated with the video image class statistical model.
-
4. A method of segmenting an audio-video recording, comprising the steps of:
-
identifying one or more video frame intervals having similarity to a predetermined video image class;
extracting one or more audio intervals corresponding to the one or more video frame intervals;
applying an acoustic clustering method on the one or more audio intervals to produce one or more audio clusters; and
wherein the step of identifying one or more video frame intervals having similarity to a predetermined video class includes the step of finding video frame intervals corresponding to slide intervals longer than a predetermined time duration. - View Dependent Claims (5, 6, 7, 8, 9, 10, 11)
parameterizing each audio interval by a mean vector; and
applying an agglomerative clustering method to euclidean distances between mean vectors corresponding to each audio interval.
-
-
6. A method as in claim 5, wherein the mean vector is a mel-frequency cepstral coefficient mean vector.
-
7. A method as in claim 6, wherein the mean vector is a filterbank or liner predictive coding coefficient mean vector.
-
8. A method as in claim 4, further comprising the steps of:
-
merging the audio intervals within same audio clusters to produced merged audio intervals; and
training source-specific speaker models on the merged audio intervals.
-
-
9. A method in claim 8, further comprising the step of:
segmenting the audio-visual recording by speaker using the source-specific speaker models to identify each speaker.
-
10. A method as in claim 8, further comprising the step of:
-
creating a speaker transition model using a speaker sequence indicated by the merged audio intervals and the source-specific speaker models; and
segmenting the audio-visual recording using the speaker transition model.
-
-
11. A method as in claim 10, wherein the speaker transition model includes a sequence of speaker units, each speaker unit including a source-specific speaker model and a filler model.
-
12. A computer readable storage medium, comprising:
-
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method of segmenting an audio-video recording, comprising the steps of;
identifying one or more video frame intervals having similarity to a predetermined video image class;
extracting one or more audio intervals corresponding to the one or more video frame intervals;
applying an acoustic clustering method on the one or more audio intervals to produce one or more audio clusters; and
wherein the step of identifying one or more video frame intervals comprises;
decimating a video portion of the audio-visual recording in time and space to produce decimated frames; and
for each decimated frame, transforming the decimated frame to produce a transform matrix;
extracting a feature vector from the transform matrix; and
determining similarity of the frame using the feature vector and a video image class statistical model. - View Dependent Claims (13)
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method as in claim 12, wherein the step of measuring similarity of the frame includes the steps of;
subtracting a video image class mean vector from the feature vector to determine a difference vector; and
comparing a magnitude of the difference vector to a predetermined multiple of a standard deviation associated with the video image class statistical model.
-
-
14. A computer readable storage medium, comprising:
-
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method of segmenting an audio-video recording, comprising the steps of;
identifying one or more video frame intervals having similarity to a predetermined video image class;
extracting one or more audio intervals corresponding to the one or more video frame intervals;
applying an acoustic clustering method on the one or more audio intervals to produce one or more audio clusters; and
wherein the step of identifying one or more video frame intervals having similarity to a predetermined video image class includes the step of finding video frame intervals corresponding to slide intervals longer than a predetermined time duration. - View Dependent Claims (15, 16, 17, 18, 19)
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method as in claim 14, wherein the step of applying an acoustic clustering method comprises the steps of;
parameterizing each audio interval by a mel-frequency cepstral coefficient mean vector; and
applying an agglomerative clustering method to euclidean distances between mel-frequency cepstral coefficient mean vectors corresponding to each audio interval.
-
-
16. A computer readable storage medium, comprising:
-
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method as in claim 14, further comprising the steps of;
merging the audio intervals within same audio clusters to produced merged audio intervals; and
training source-specific speaker models on the merged audio intervals.
-
-
17. A computer readable storage medium, comprising:
-
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method as in claim 16, further comprising the step of;
segmenting the audio-visual recording by speaker using the source-specific speaker models to identify each speaker.
-
-
18. A computer readable storage medium, comprising:
-
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method as in claim 16, further comprising the step of;
creating a speaker transition model using a speaker sequence indicated by the merged audio intervals and the source-specific speaker models; and
segmenting the audio-visual recording using the speaker transition model.
-
-
19. A computer readable storage medium, comprising:
computer readable program code embodied on said computer readable storage medium, said computer readable program code for programming a computer to perform a method as in claim 18, wherein the speaker transition model includes a sequence of speaker units, each speaker unit including a source-specific speaker model and a filler model.
Specification