Using codec parameters for endpoint detection in speech recognition
AssigneeNuance Communications, Inc.
InventorsMichael D. Edgington, Stephen W. Laverty, Gunnar Evermann
Life of this patent
Estimating the term…
Abstract
Systems, methods and apparatus for determining an estimated endpoint of human speech in a sound wave received by a mobile device having a speech encoder for encoding the sound wave to produce an encoded representation of the sound wave. The estimated endpoint may be determined by analyzing information available from the speech encoder, without analyzing the sound wave directly and without producing a decoded representation of the sound wave. The encoded representation of the sound wave may be transmitted to a remote server for speech recognition processing, along with an indication of the estimated endpoint.
Claims50
A method for use in a system comprising a first device that receives human speech and a second device that comprises a speech recognizer, wherein the first device receives at least one sound wave comprising the human speech and encodes the received at least one sound wave, via at least one speech encoder in the first device, to produce at least one encoded representation of the at least one sound wave, wherein the first device transmits the at least one encoded representation to the second device, wherein the second device decodes the at least one encoded representation and performs, via the speech recognizer, speech recognition on the human speech, the method, performed by the first device, comprising acts of:
- determining an estimated endpoint of the human speech in the at least one sound wave by analyzing information available from the at least one speech encoder, without analyzing the at least one sound wave and without producing a decoded representation of the at least one sound wave;
- and
- providing to the second device an indication of the estimated endpoint.
At least one non-transitory computer readable medium having encoded thereon instructions that, when executed by at least one processor, perform a method for use in a system comprising a first device that receives human speech and a second device that comprises a speech recognizer, wherein the first device receives at least one sound wave comprising the human speech and encodes the received at least one sound wave, via at least one speech encoder in the first device, to produce at least one encoded representation of the at least one sound wave, wherein the first device transmits the at least one encoded representation to the second device, wherein the second device decodes the at least one encoded representation and performs, via the speech recognizer, speech recognition on the human speech, the method, performed by the first device, comprising acts of:
- determining an estimated endpoint of the human speech in the at least one sound wave by analyzing information available from the at least one speech encoder, without analyzing the at least one sound wave and without producing a decoded representation of the at least one sound wave;
- and
- providing to the second device an indication of the estimated endpoint.
A first device for use in a system comprising at least one second device configured to decode at least one encoded representation of at least one sound wave comprising human speech, the at least one second device comprising a speech recognizer to perform speech recognition on the human speech, the first device comprising:
- at least one speech encoder to encode the at least one sound wave to produce the at least one encoded representation, the at least one sound wave being received at the first device;
- at least one endpoint detection circuit to determine an estimated endpoint of the human speech in the at least one sound wave by analyzing information available from the at least one speech encoder, without analyzing the at least one sound wave and without producing a decoded representation of the at least one sound wave;
- and
- at least one transmitter to transmit the at least one encoded representation of the at least one sound wave and an indication of the estimated endpoint to the at least one second device.
Dependencies are inferred from the claim text.