Back to results
US 6,275,800 B1Granted
Voice recognition system and method
AssigneeMotorola, Inc.
InventorsDavid Erik Chevalier, Henry L. Kazecki
Life of this patent
Estimating the term…
Abstract
A voice recognition system (204, 206, 207, 208) generates a variable rejection strictness as a function of at least one background noise level measured during training and noise signal measurements made during an input utterance made during recognition mode of operation. A word entrance penalty is assigned as a function of the variable rejection strictness.
Claims10
A method of operating a voice recognition system, comprising the steps of:
- calculating a noise reference feature, in a training mode, as a function of a training input utterance;
- calculating an input noise value, in a recognition mode, as a function of an input utterance;
- calculating a reject value as a ratio of the input noise value and the noise reference feature;
- and
- determining a best match and selecting the determined best match in response to the determined best match being greater than the reject value.
2 more elements
A method of operating a voice recognition system, comprising the steps of:
- generating a variable rejection strictness as a function of at least one background noise level measured during training and noise signal measurements made during an input utterance made during recognition mode of operation;
- and
- deriving a word entrance penalty as a function of the variable rejection strictness, wherein during recognition, the input noise signal measurements are compared to the measured background noise level and a noise ratio is computed, and wherein the voice recognition system determines a best match and selects the determined best match using the word entrance penalty.
A radiotelephone comprising:
- a microphone for inputting user utterances;
- and
- voice recognition system coupled to the microphone, the voice recognition system calculating a noise reference feature, in a training mode, as a function of a training input utterance, calculating an input noise value, in a recognition mode, as a function of an input utterance, calculating a reject value as a ratio of the input noise value and the noise reference feature, wherein the voice recognition system determines a best match and selects the determined best match in response to the determined best match being greater than the reject value.
Dependencies are inferred from the claim text.