Speech recognition system

• US 4,239,936 A
• Filed: 12/28/1978
• Issued: 12/16/1980
• Est. Priority Date: 12/28/1977
• Status: Expired due to Term

• Current Assignee: Nippon Electric Company Limited (NEC Corporation)
• Original Assignee: Nippon Electric Company Limited (NEC Corporation)
• Inventors: Sakoe, Hiroaki
• Primary Examiner(s): Morrison, Malcolm A.
• Assistant Examiner(s): Kemeny, E. S.
• Application Number:
  US05/974,274
• Time in Patent Office: 719 Days
• Field of Search: 179/1 SD, 179/1 P, 179/1 VC, 179/1 VL
• US Class Current: 704/233
• CPC Class Codes: G10L 15/20 Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech G10L21/02 takes precedence
• CPC Class Codes: G10L 25/87 Detection of discrete points within a voice signal

Abstract

A speech recognition system adaptable to noisy environments is disclosed. The system includes a recognition unit for recognizing input speech signals and a noise measuring unit for measuring the intensity of ambient noises. The system also includes a rejection unit responsive to a rejection standard controlled by the intensity of the measured noise for rejecting the rejection results given from the recognition unit when the rejection standard is exceeded.

Claims

1. A speech recognition system comprising:

a first microphone for picking up input speech signals;

a speech-signal-duration determining unit for determining speech signal durations by monitoring the amplitude of the input speech signals;

a recognition unit for determining recognition results by analyzing and recognizing signals given in said speech signal durations;

a second microphone provided to pick up ambient noises;

means for measuring the intensity of input signals given from said second microphone; and

a rejection unit which, when the intensity of said noise exceeding a predetermined standard value is detected in any of said speech signal durations, rejects the recognition results determined by said recognition unit.

2. A speech recognition system as recited in claim 1, wherein said recognition unit comprises:

analyzer means for receiving said input speech signals and generating a digital output representing a time series of spectrum vectors;

buffer means receiving the digital output of said analyzer means for temporarily storing said digital output in response to an output from said speech-signal-duration determining unit;

reference pattern memory means for digitally storing reference patterns represented by a plurality of time series of spectrum vectors; and

means for matching the contents of said buffer means with one of the reference patterns stored in said reference pattern memory means and generating a recognition output and a similarity measure signal.

3. A speech recognition system as recited in claim 2, wherein said noise-intensity measuring means comprises:

a rectifier connected to receive input signals from said second microphone;

resettable integrator means for integrating an input signal;

switch means responsive to the output of said speech-signal-duration determining unit for connecting the output of said rectifier to the input of said resettable integrator means; and

means responsive to a rise in the output of said speech-signal-determining unit for resetting said resettable integrator means.

4. A speech recognition system as recited in claim 2, wherein said noise-intensity measuring means comprises:

converter means connected to receive input signals from said second microphone for providing a digital output;

accumulator means connected to receive the digital output from said converter means and integrate successive digital outputs by accumulating the same;

register means connected to receive and temporarily store a digital output from said converter means;

comparator means for comparing the outputs of said converter means and said register means to detect the maximum intensity of noise in speech signal durations and generate an output signal to said register means to control the input of digital signals thereto;

means responsive to a rise in the output of said speech-signal-determining unit for resetting both said accumulator means and said register means; and

evaluation means responsive to the contents of said accumulator means and said register means for an output signal representing ambient noise intensity.

5. A speech recognition system as recited in claim 1, wherein said rejection unit comprises:

threshold circuit means for comparing the output of said noise-intensity measuring means with a preset threshold and generating a gating signal; and

gate means responsive to said gating signal for selectively rejecting the output of said recognition unit.

6. A speech recognition system comprising:

a speech microphone provided to pick up input speech signals;

means for calculating the amplitude of the speech signals picked up by said speech microphone;

a noise microphone provided to pick up ambient noises;

means for calculating the intensity of the noises picked up by said noise microphone;

a speech-signal detecting unit for determining speech signal durations by monitoring the amplitude of said speech signals compensated for by the intensity of said noises;

a recognition unit for recognizing said input speech signals given in speech signal durations; and

a rejection unit which, when the intensity of said noise exceeding a predetermined standard value is detected in any of said speech signal durations, rejects the recognition results determined by said recognition unit.

7. A speech recognition system comprising:

a first microphone for picking up input speech signals;

a speech-signal-duration determining unit for determining speech signal durations by monitoring the amplitude of the input speech signals;

a recognition unit for determining recognition results by analyzing and recognizing signals given in said speech signal durations, having analyzer means for receiving said input speech signals and generating a digital output representing a time series of spectrum vectors, buffer means receiving the digital output of said analyzer means for temporarily storing said digital output in response to an output from said speech-signal-duration determining unit, reference pattern memory means for digitally storing reference patterns represented by a plurality of time series of spectrum vectors, and matching the contents of said buffer means with one of the reference patterns stored in said reference pattern memory means and generating a recognition output and a similarity measure signals;

a second microphone provided to pick up ambient noises;

means for measuring the intensity of input signals given from said second microphone; and

a rejection unit for rejecting the recognition results by said recognition unit, having threshold control means for setting a threshold value in response to the output of said noise intensity measuring means, comparator means receiving said similarity measure signal from said matching means and comparing said similarity measure signal with said threshold value for generating a gating signal, and gate means responsive to said gating signal for selectively passing said recognition output from said matching means.