Method for formation of speech recognition parameters
AssigneeMotorola, Inc.
InventorsDusan Macho, Yan Ming Cheng
Life of this patent
Estimating the term…
Abstract
A voice sample characterization front-end suitable for use in a distributed speech recognition context. A digitized voice sample 31 is split between a low frequency path 32 and a high frequency path 33. Both paths are used to determine spectral content suitable for use when determining speech recognition parameters (such as cepstral coefficients) that characterize the speech sample for recognition purposes. The low frequency path 32 has a thorough noise reduction capability. In one embodiment, the results of this noise reduction are used by the high frequency path 33 to aid in de-noising without requiring the same level of resource capacity as used by the low frequency path 32.
Claims21
A method comprising:
- providing a digitized voice signal;
- using the digitized voice signal to provide:
- a first signal comprising only a first portion of the digitized voice signal;
- and
- a second signal comprising only a second portion of the digitized voice signal, wherein the first portion and the second portion of the digitized voice signal are at least partially different from one another;
- using the first signal to form a first set of spectral information;
- using the second signal to form a second set of spectral information;
- forming speech recognition parameters based upon both the first and second set of spectral information, wherein the speech recognition parameters correspond to the digitized voice signal.
A method comprising:
- providing a digitized voice signal;
- using the digitized voice signal to provide:
- a first signal comprising only a first portion of the digitized voice signal;
- and
- a second signal comprising only a second portion of the digitized voice signal, wherein the first portion and the second portion of the digitized voice signal are at least partially different from one another;
- processing the first signal to reduce at least some noise as may be present in the first signal to provide a noise reduced first signal;
- processing the second signal to reduce at least some noise as may be present in the second signal to provide a noise reduced second signal;
- using the first noise reduced signal to form a first set of spectral information;
- using the second noise reduced signal to form a second set of spectral information;
- forming speech recognition parameters based upon both the first and second set of spectral information, wherein the speech recognition parameters correspond to the digitized voice signal.
A method comprising:
- providing a digitized voice signal;
- low pass filtering the digitized voice signal to provide a first digitized voice signal;
- high pass filtering the digitized voice signal to provide a second digitized voice signal;
- processing the first digitized voice signal using a first process to:
- provide a corresponding spectral estimation;
- and
- to reduce at least some noise as may be present in the first digitized voice signal to provide a noise reduced first digitized voice signal;
- providing a processed signal based on the second digitized voice signal;
- processing the processed signal as a function, at least in part, of the corresponding spectral estimation to provide a coded representation;
- processing the coded representation as a function, at least in part, of the noise reduced first digitized voice signal to provide a de-coded representation;
- processing a signal that is based upon the second digitized voice signal as a function, at least in part, of the de-coded representation to reduce at least some noise as may be present in the signal and to provide resultant high frequency information as corresponds to the digitized voice signal;
- combining the resultant high frequency information with low frequency information as corresponds to the digitized voice signal to provide combined spectral information for the digitized voice signal;
- forming speech recognition parameters based upon the combined spectral information, wherein the speech recognition parameters correspond to the digitized voice signal.
Dependencies are inferred from the claim text.