Systems and methods for implementing segmentation in speech recognition systems
AssigneesVerizon Corporate Services Group Inc., BBN Technologies Corp.
InventorsChang-Qing Shu, Han Shu
Life of this patent
Estimating the term…
Abstract
A speech recognition system (105) includes an acoustic front end (115) and a processing unit (125). The acoustic front end (115) receives frames of acoustic data and determines cepstral coefficients for each of the received frames. The processing unit (125) determines a number of peaks in the cepstral coefficients for each of the received frames of acoustic data and compares the peaks in the cepstral coefficients of a first one of the received frames with the peaks in the cepstral coefficients of at least a second one of the received frames. The processing unit (125) then segments the received frames of acoustic data based on the comparison.
Claims16
A method of segmenting acoustic data for use in a speech recognition process, comprising:
- receiving frames of acoustic data;
- determining cepstral coefficients for each of the received frames of acoustic data;
- determining how many local maxima there are in a plot of the cepstral coefficients for each received frame of acoustic data;
- and
- segmenting the received frames of acoustic data based on results of said local maxima determining.
A computerized speech recognition system, comprising:
- an acoustic front end configured to:
- receive frames of acoustic data,
- determine cepstral coefficients for each of the received frames of acoustic data;
- and
- a processing unit configured to:
- determine how many local maxima there are in a plot of the cepstral coefficients for each of the received frames of acoustic data,
- compare said how many local maxima there are in the plot of the cepstral coefficients of a first one of the received frames with said how many local maxima there are in the plot of the cepstral coefficients of at least a second one of the received frames, and
- segment the received frames of acoustic data based on the comparison.
A computer-readable medium excluding carrier waves, said medium containing instructions for controlling at least one processing unit to perform a method of segmenting acoustic data for use in a speech recognition process, the method comprising:
- receiving cepstral coefficients corresponding to frames of acoustic data;
- segmenting the frames of acoustic data based on the received cepstral coefficients;
- and
- determining how many local maxima there are in a plot of the cepstral coefficients corresponding to each of the frames of acoustic data, wherein the segmenting of the frames of acoustic data is further based on the determining.
A method of recognizing patterns in acoustic data, comprising:
- receiving frames of acoustic data;
- determining cepstral coefficients corresponding to the received frames of acoustic data;
- determining a number of peaks in a plot of the cepstral coefficients for each received frame of acoustic data;
- determining at least one weighting parameter based on the determined number of peaks in the plot of the cepstral coefficients;
- and
- recognizing patterns in the received frames of acoustic data using the at least one weighting parameter.
A speech recognition system, comprising:
- an acoustic front end configured to receive frames of acoustic data and determine a number of peaks in a plot of cepstral coefficients for each of the received frames,
- a processing unit configured to:
- determine segmentation information corresponding to the determined number of peaks,
- determine at least one weighting parameter based on the determined segmentation information, and
- recognize patterns in the received frames of acoustic data using the at least one weighting parameter.
A computer readable medium excluding carrier waves, said medium having encoded thereon a data structure comprising:
- cepstral coefficient data corresponding to each frame of a plurality of frames of acoustic data, the cepstral coefficient data including how many local maxima there are in a plot of cepstral coefficients corresponding to each frame of acoustic data;
- and
- segmentation data indicating segmentation of the frames of acoustic data based on said how many local maxima there are.
An acoustic recognition system, comprising:
- an acoustic front end configured to:
- receive frames of acoustic data and determine how many local maxima there are in a plot of cepstral coefficients for each of the received frames;
- a processing unit configured to generate, based on the determined how many local maxima there are, end frame numbers for each phoneme or Hidden Markov Model (HMM) state contained in the received frames;
- and
- a trainer/HMM decoder configured to use the generated end frame numbers for determining weighted scores that can be used for recognition of acoustic events contained in the received frames of acoustic data.
An acoustic recognition system, comprising:
- an acoustic front end configured to receive frames of acoustic data and determine a number of peaks of cepstral coefficients for each of the received frames;
- and
- a processing unit configured to:
- generate, based on the determined number of peaks, end frame numbers for each phoneme or Hidden Markov Model (HMM) state contained in the received frames of acoustic data, and
- determine weighted scores based on the generated end frame numbers that can be used for recognition of acoustic events contained in the received frames of acoustic data.
Dependencies are inferred from the claim text.