Method and apparatus for recognizing a speaker in lawful interception systems
First Claim
1. A method for associating a voice of a first human speaker with at least one of a multiplicity of speakers, each of the multiplicity of speakers associated with an acoustic model and with data, the method comprising the steps of:
- capturing an audio signal of the first human speaker communicating with a second human speaker and extracting the voice of the first human speaker;
receiving or extracting the data associated with each of the multiplicity of speakers;
generating tags associated with each of the acoustic models of the multiplicity of speakers;
wherein the tags represent an at least one parameter associated with the acoustic model of the speaker and with at least one parameter associated with the acoustic models of persons who are known to be connected directly or indirectly via one or more middle persons to a known speaker in the interaction;
constructing one or more groups comprising acoustic models of speakers based on tag value similarity;
determining an order for the groups by giving high priority to groups with persons who have direct connection with the first speaker or the second speaker and lower priority to groups with persons having indirect connection to the first speaker or the second speaker;
selecting at least one group to be matched against the voice of the first speaker;
determining an at least one non-acoustic score between data related to the first speaker, and the tags of the at least one matched group or the tags of an at least one acoustic model from the at least one matched group;
determining an at least one acoustic score between the voice of the first speaker and an at least one acoustic model from the at least one matched group;
obtaining a total score by combining the non-acoustic score with the acoustic score;
determining according to the total score whether an identification criteria was met; and
if the identification criteria was met, associating the first speaker with the at least one model from the matched group.
2 Assignments
0 Petitions
Accused Products
Abstract
A method and apparatus for identifying a speaker within a captured audio signal from a collection of known speakers. The method and apparatus receive or generate voice representations for each known speakers and tag the representations according to meta data related to the known speaker or to the voice. The representations are grouped into one or more groups according to the indices. When a voice to be recognized is introduced, characteristics are determined according to which the groups are prioritized, so that the representations participating only in part of the groups are matched against the voice to be identified, thus reducing identification time and improving the statistical significance.
-
Citations
32 Claims
-
1. A method for associating a voice of a first human speaker with at least one of a multiplicity of speakers, each of the multiplicity of speakers associated with an acoustic model and with data, the method comprising the steps of:
-
capturing an audio signal of the first human speaker communicating with a second human speaker and extracting the voice of the first human speaker; receiving or extracting the data associated with each of the multiplicity of speakers; generating tags associated with each of the acoustic models of the multiplicity of speakers;
wherein the tags represent an at least one parameter associated with the acoustic model of the speaker and with at least one parameter associated with the acoustic models of persons who are known to be connected directly or indirectly via one or more middle persons to a known speaker in the interaction;constructing one or more groups comprising acoustic models of speakers based on tag value similarity; determining an order for the groups by giving high priority to groups with persons who have direct connection with the first speaker or the second speaker and lower priority to groups with persons having indirect connection to the first speaker or the second speaker; selecting at least one group to be matched against the voice of the first speaker; determining an at least one non-acoustic score between data related to the first speaker, and the tags of the at least one matched group or the tags of an at least one acoustic model from the at least one matched group; determining an at least one acoustic score between the voice of the first speaker and an at least one acoustic model from the at least one matched group; obtaining a total score by combining the non-acoustic score with the acoustic score; determining according to the total score whether an identification criteria was met; and if the identification criteria was met, associating the first speaker with the at least one model from the matched group. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14)
-
-
15. An apparatus for associating a voice of a first human speaker, with at least one of a multiplicity of speakers, each of the multiplicity of speakers associated with an acoustic model and with data, the apparatus comprising:
-
a storage device for storing the acoustic model and associated meta data; a capturing or logging component for receiving a voice sample of the first human speaker extracted from a captured audio signal of a communication between the first human speaker and a second human speaker; a tagging component for generating tags associated with the acoustic models of the multiplicity of speakers;
wherein the tags represent an at least one parameter associated with the acoustic model of the speaker and with at least one parameter associated with the acoustic models of persons who are known to be connected directly or indirectly via one or more middle persons to a known speaker in the interaction; andfor constructing one or more groups comprising acoustic models of speakers based on tag value similarity; a selection component for determining an order for the groups by giving high priority to groups with persons who have direct connection with the first speaker or the second speaker and lower priority to groups with persons having indirect connection to the first speaker or the second speaker; and
selecting a group comprising an at least one acoustic model for matching with the voice sample of the first speaker;a non-acoustic score determination component, for determining a non-acoustic score between data related to the first speaker, and the tags of the at least one matched group or the tags of an at least one acoustic model from the at least one matched group; an acoustic score determination component for determining an acoustic score between the voice of the first speaker and an at least one acoustic model from the at least one matched group; a combining component for combining the acoustic score and the non-acoustic score into a total score; and a criteria evaluation component for determining whether the total score meets an at least one criteria. - View Dependent Claims (16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30)
-
-
31. A method for associating a voice of a first human speaker with at least one of a multiplicity of speakers, each of the multiplicity of speakers associated with an acoustic model and with data, the method comprising the steps of:
-
capturing an audio signal of the first human speaker communicating with a second human speaker and extracting the voice of the first human speaker and the second human speaker; receiving or extracting the data associated with each of the multiplicity of speakers; generating tags associated with each of the acoustic models of the multiplicity of speakers;
wherein the tags represent an at least one parameter associated with the acoustic model of the speaker and with at least one parameter associated with the acoustic models of persons who are known to be connected directly or indirectly via one or more middle persons to a known speaker in the interaction;constructing one or more groups comprising acoustic models of speakers based on tag value similarity; determining an order for the groups by giving high priority to groups with persons who have direct connection with the first speaker or the second speaker and lower priority to groups with persons having indirect connection to the first speaker or the second speaker; selecting at least one group to be matched against the voice of the first speaker or voice of the second speaker according to the group order; and associating the first speaker or the second speaker with the at least one model from the matched group. - View Dependent Claims (32)
-
Specification