Lombard voice recognition method and apparatus for recognizing voices in noisy circumstance
AssigneeSharp Kabushiki Kaisha
InventorShin Kamiya
Life of this patent
Estimating the term…
Abstract
A Lombard voice recognition method for recognizing a voice input in a noisy background includes a step of matching and a step of warping. The step of matching matches a frequency spectrum of a feature pattern derived from the input voice to a frequency spectrum of a standard pattern. The step of warping warps a lower frequency spectrum of the feature pattern than a lower frequency side with respect to a frequency axis in case that a noise level of the background in higher than a predetermined noise level.
Claims18
A Lombard voice recognition method for recognizing an input voice under a noisy background comprising the steps of:
- warping a frequency spectrum of a lower frequency than a predetermined frequency in a frequency spectrum of a feature pattern with a predetermined warping width toward a lower frequency side in a frequency axis, said warping being at a time when a noisy level of an input voice is higher than a predetermined noisy level, said frequency spectrum of said feature pattern being derived from said input voice;
- and
- matching a frequency spectrum of a standard pattern with said warped frequency spectrum of said feature pattern to calculate similarity between said standard pattern and said feature pattern and outputting a recognized result based on the calculated similarity,
- said predetermined warping width being changeable in accordance with a nose level of said input voice at a time when said lower frequency spectrum of said feature pattern than said predetermined frequency is warped by said predetermined warping width toward said lower frequency side with respect to said frequency axis.
A Lombard voice recognition apparatus which is capable of recognizing a voice input in a noisy background, said apparatus having a control means for controlling said apparatus in a proper manner, said apparatus comprising:
- means connected to said control means for extracting a power spectrum from a voice of a particular speaker;
- means connected to said extracting means for cutting said power spectrum at proper syllabic intervals and for outputting a feature pattern;
- means connected to said cutting means for calculating similarity between said feature pattern and a standard pattern, and for outputting recognized result based on the calculated similarity;
- means connected to said extracting means for detecting a noise level of said noisy background;
- means connected to said detecting means and said similarity calculating means for warping said power spectrum located in a lower frequency region than a predetermined value toward a lower frequency side;
- and
- means connected to said similarity calculating means for storing said standard pattern.
Dependencies are inferred from the claim text.