초록 ▼

A method for identifying a plurality of speakers in audio data and for decoding the speech spoken by said speakers; the method comprising:receiving speech;dividing the speech into segments as it is received;processing the received speech segment by segment in the order received to identify the speak

A method for identifying a plurality of speakers in audio data and for decoding the speech spoken by said speakers; the method comprising:receiving speech;dividing the speech into segments as it is received;processing the received speech segment by segment in the order received to identify the speaker and to decode the speech, processing comprising: performing primary decoding of the segment using an acoustic model and a language model;obtaining segment parameters indicating the differences between the speaker of the segment and a base speaker during the primary decoding;comparing the segment parameters with a plurality of stored speaker profiles to determine the identity of the speaker, and selecting a speaker profile for said speaker;updating the selected speaker profile;performing a further decoding of the segment using a speaker independent acoustic model, adapted using the updated speaker profile;outputting the decoded speech for the identified speaker, wherein the speaker profiles are updated as further segments of speech relating to a speaker profile are processed.

대표청구항 ▼

1. A method for identifying a plurality of speakers in audio data and for decoding the speech spoken by said speakers; the method comprising:receiving speech;dividing the speech into segments as it is received;processing the received speech segment by segment in the order received to identify the sp

1. A method for identifying a plurality of speakers in audio data and for decoding the speech spoken by said speakers; the method comprising:receiving speech;dividing the speech into segments as it is received;processing the received speech segment by segment in the order received to identify the speaker and to decode the speech, processing comprising: performing primary decoding of the segment using an acoustic model and a language model;obtaining segment parameters indicating the differences between the speaker of the segment and a base speaker during the primary decoding, comprising obtaining the parameters which allow a speaker transform to be estimated, said speaker transform adapting the speech of the speaker of the segment to that of the base speaker of the acoustic model;comparing the segment parameters with a plurality of stored speaker profiles to determine the identity of the speaker, and selecting a speaker profile for said speaker;updating the selected speaker profile using said segment parameters;performing a further decoding of the segment using a speaker independent acoustic model, adapted using the updated speaker profile;outputting the decoded speech for the identified speaker, wherein the speaker profiles are updated as further segments of speech relating to a speaker profile are processed. 2. A method according to claim 1, wherein if the segment parameters do not match closely enough with a stored speaker profile, a new speaker profile is initialised based on the parameters obtained for the segment. 3. A method according to claim 2, wherein a new speaker profile is initialised if the likelihood or value of auxiliary function is greater for a unity transform or generic transform than for one of the stored speaker profiles. 4. A method according to claim 2, wherein a new speaker profile is initialised if the likelihood or value of auxiliary function is less than a predetermined threshold. 5. A method according to claim 1, wherein the speaker transform is a MLLR or CMLLR transform. 6. A method according to claim 1, wherein the speaker profile comprises both adaptive and prior statistics. 7. A method according to claim 1, wherein the primary decoding uses a language model. 8. A method according to claim 1, wherein secondary decoding comprises rescoring a lattice of possible text corresponding to the segment. 9. A method according to claim 1, wherein dividing the input speech into segments comprises detecting where there is silence in the input speech. 10. A method according to claim 1, wherein the base speaker is a canonical speaker. 11. A method according to claim 1, wherein the base speaker is the speaker of a previous segment. 12. A non-transitory carrier medium carrying computer readable instructions for controlling the computer to carry out the method of claim 1. 13. A system for identifying a plurality of speakers in audio data and for decoding the speech spoken by said speakers; the system comprising: a receive for audio containing speech; anda processor, said processor being adapted to: divide the speech into segments as it is received;process the received speech segment by segment in the order received to identify the speaker and to decode the speech, processing comprising:perform primary decoding of the segment using an acoustic model and a language model;obtain segment parameters indicating the differences between the speaker of the segment and a base speaker during the primary decoding,wherein obtaining segment parameters comprises obtaining the parameters which allow a speaker transform to be estimated, said speaker transform adapting the speech of the speaker of the segment to that of the base speaker of the acoustic model;compare the segment parameters with a plurality of stored speaker profiles to determine the identity of the speaker, and selecting a speaker profile for said speaker;update the selected speaker profile using said segment parameters; andperform a further decoding of the segment using a speaker independent acoustic model, adapted using the updated speaker profile,the system further comprising an output for outputting the decoded speech for the identified speaker,wherein the speaker profiles are updated as further segments of speech relating to a speaker profile are processed.