Method and apparatus for mitigating the effect of transmission errors in a distributed speech recognition process and system
View Patent ↗A method and apparatus for distributed speech recognition serve to mitigate the effect of transmission errors. The method comprises the steps of identifying speech recognition parameters which have been subjected to a transmission error, and processing data to be sent to the speech recognition decoder to ensure that any speech recognition parameters which have been subjected to such an error are excluded from back-end processing. The speech recognition parameters which have been subjected to transmission errors are excluded from back-end processing by replacing those parameters with data which is selected so as to be rejected by the speech recognition decoder as abnormal and/or non-speechlike.
1. A method of mitigating the effect of transmission errors in a distributed speech recognition process, the distributed speech recognition process being one in which speech recognition parameters are arranged in vectors corresponding to sampling time-frames at a first location, and are received at a second location for processing by a speech recognition decoder having been transmitted from the first location, the method comprising the steps of:
receiving one or more vectors, each vector including one or more speech recognition parameters;
identifying a group comprising one or more of said vectors which has undergone a transmission error; and
processing data to be sent to the speech recognition decoder to ensure that at least one speech recognition parameter in the identified group is excluded from backend processing.
2. A method according to claim 1 in which all the speech recognition parameters of each vector of said group are excluded from backend processing.
3. A method according to claim 1 in which the step of processing data to be sent to the speech recognition decoder comprises the step of replacing said at least one speech recognition parameter with data selected so as to be rejected by the speech recognition decoder as abnormal.
4. A method according to claim 3 in which the selected data comprises one or more speech recognition parameter having a value substantially different from any that would occur in natural speech.
5. A method according to claim 3 in which the speech recognition parameters are subjected to quantization before transmission, and reconstitution after transmission, wherein the quantization bounds a range of possible values which the parameters can take on reconstitution and said selected data comprises one or more value outside said range.
6. A method according to claim 5 further comprising the steps of: using a split vector quantization technique in which each quantizer operates on a pair of
speech recognition parameters; and choosing said selected data to comprise one or more value removed from each quantizer
centroid by at least a predetermined amount.
7. A method according to claim 3 in which the backend processing comprises the step of substituting a default value for the score generated when processing the selected abnormal data.
8. An apparatus for mitigating the effect of transmission errors in a distributed speech recognition process, the distributed speech recognition process being one in which speech recognition parameters are arranged in vectors corresponding to sampling time-frames at a first location and are received at a second location for processing by a speech recognition decoder having been transmitted from the first location, the apparatus comprising: receiving means for receiving one or more vectors, each vector including one or more speech recognition parameters;
identifying means for identifying a group comprising one or more of said vectors which has undergone a transmission error; and
data processing means for processing data to be sent to the speech recognition decoder to ensure that at least one speech recognition parameter in the identified group is excluded from backend processing.
9. An apparatus according to claim 8 in which the processing means is arranged to exclude all the speech recognition parameters of each vector of said group from backend processing.
10. An apparatus according to claim 8 in which the processing means is arranged to replace said at least one speech recognition parameter with data selected so as to be rejected by the speech recognition decoder as abnormal.
11. An apparatus according to claim 10 in which the selected data comprises one or more speech recognition parameter having a value substantially different from any that would occur in natural speech.
12. An apparatus according to claim 10 further comprising: means for quantizing the speech recognition parameters before transmission, and means for reconstitution of the parameters after transmission, wherein the quantization
bounds a range of possible values which the parameters can take on reconstitution and said selected data comprises one or more value outside said range.
13. An apparatus according to claim 12 in which the quantization means is arranged for performing a split vector quantization in which each quantizer operates on a pair of speech recognition parameters and said selected data comprises one or more value removed, by at least a predetermined amount, from each quantizer centroid.
14. An apparatus according to claim 8 in which the backend decoder comprises means for substituting a default value for the score generated when processing the selected abnormal data.