Pre-processing system for speech recognition

• US 4,227,046 A
• Filed: 02/24/1978
• Issued: 10/07/1980
• Est. Priority Date: 02/25/1977
• Status: Expired due to Term

• Current Assignee: Hitachi, Ltd.
• Original Assignee: Hitachi, Ltd.
• Inventors: Ichikawa, Akira, Nakajima, Akira
• Primary Examiner(s): Morrison, Malcolm A.
• Assistant Examiner(s): Kemeny, E. S.
• Application Number:
  US05/880,951
• Time in Patent Office: 956 Days
• Field of Search: 179/1 SA, 179/1 SD, 179/1 D, 179/1 P
• US Class Current: 704/228
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• 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

Abstract

A pre-processing system for a speech recognition system nullifies the distortion in a received signal caused by the transmission line by forming a normalized received signal whereby the distortion is cancelled: the received signal is spectrally averaged to form a mean spectrum which is used to adjust a filter to a characteristic inverse to the distortion; this filter then filters the received signal to form the normalized received signal s_j '"'"'(ω,t)=S_j (ω,t)/S_j (ω) which is equal to the normalized source signal =S_i (ω,t)/S_i (ω).

Claims

1. A pre-processing system for speech recognition comprising:

first means for receiving input speech which has been transmitted through a transmission medium, said first means providing an output representative of said input speech;

second means connected to the output of said first means for determining the mean frequency spectrum of the input speech received by said first means and providing an output representative of said mean frequency spectrum; and

filter means, connected to the outputs of said first and seconds means, having a frequency spectrum response inversely proportional to the mean frequency spectrum determined by said second means, for filtering the input signal received by said first means.

2. A pre-processing system for speech recognition according to claim 1, which further comprises third means connected between said first and second means, for determining a time boundary for analyzing said frequency spectrum of the input speech on the basis of the input speech received by said first means and for inputting the input speech in the determined time boundary to said second means.

3. A pre-processing system for speech recognition according to claim 2, wherein said third means determines the time boundary for analyzing said frequency spectrum of the input speech by determining the start and end points of said input speech.

4. A pre-processing system for speech recognition according to claim 1, wherein said first means comprises conversion means for converting the input speech into a digital signal, and storage means for storing the input speech converted by said conversion means.

5. A pre-processing system for speech recognition according to claim 1, wherein said second means comprises means for extracting partial auto-correlation coefficients of first to N-th (N being a positive integer) orders on the basis of the input speech received by said first means.

6. A pre-processing system for speech recognition according to claim 2, wherein said second means comprises means for extracting partial auto-correlation coefficients of first to N-th (N being a positive integer) orders on the basis of the input speech received by said first means.

7. A pre-processing system for speech recognition according to claim 5, wherein said second means comprises extraction means for extracting the partial auto-correlation coefficients of first to N-th orders from each of plural portions of the input speech, and calculation means for calculating a means value of the partial auto-correlation coefficients extracted by said extraction means.

8. A pre-processing system for speech recognition according to claim 6, wherein said filter means includes a holding means into which is set the mean partial auto-correlation coefficients calculated by said calculation means.

9. A pre-processing system for speech recognition according to claim 5, wherein said extracting means comprises a delay circuit for delaying the input speech received by said first means and a correlator for evaluating the correlation between the input speech delayed by said delay circuit and the input speech received by said first means and for obtaining as an output the partial auto-correlation coefficient, first and second calculation means for obtaining frontward an backward preestimated waveform values, respectively, and third and fourth calculation means for obtaining frontward and backward preestimated signal differences, respectively.

10. A pre-processing system for speech recognition according to claim 5, wherein said extracting means comprises:

a first delay circuit for delaying the input speech received by said first means;

a first correlator for evaluating the correlation between the input speech delayed by said first delay circuit and the input speech received by said first means and for obtaining as a first coefficient output the PARCOR coefficient of the first order;

first frontward and backward multipliers for multiplying the input speech from said first delay circuit and the input speech from said first correlator and obtaining frontward and backward preestimated waveform values of the first order, respectively;

a first frontward adder for calculating the difference between the delayed input speech from said first delay circuit and the backward preestimated waveform value from said first backward multiplier and for obtaining frontward preestimated signal difference of the first order;

a first backward adder for calculating the difference between the input speech from said first means and the frontward preestimated waveform value from said first frontward multiplier and for obtaining backward preestimated signal difference of the first order;

second to Nth correlators for evaluating the correlation between signals from said first to N-1th frontward adders, respectively;

second to Nth correlators for evaluating the correlation between signals from said second to Nth delay circuits and signals from said first to N-1th coefficient outputs the PARCOR coefficients of the second to N-th order, respectively;

second to N-th frontward multipliers for multiplying signals from said second to N-th delay circuits by signals from said second to N-th correlators, respectively;

second to N-th backward multipliers for multiplying signals from said first to N-th backward adders by signals from said second to Nth correlators, respectively;

second to N-th frontward adders for obtaining difference between signals from said second to N-th delay circuits and signals from said second to N-th backward multipliers, respectively; and

second to Nth backward adders for obtaining the difference between signals from said second to N-th frontward multipliers and signals from said first to N-1th backward adders.

11. A pre-processing system for speech recognition comprising:

first means for receiving an input speech signal which is a product of true speech and a distortion signal produced by transmission characteristics of a transmission system over which said speech is transmitted, wherein said first means provides an output representative of said received input speech signal;

second means connected to the output of said first means for determining the mean frequency spectrum of the received input speech signal and providing an output representative of said mean frequency spectrum; and

filter means, connected to the outputs of said first and second means, for using the mean frequency spectrum determined by the second means to provide a frequency response inversely proportional to the mean frequency spectrum determined by said second means, and for filtering the signal received from the first means representing the received input speech signal with the inversely proportional frequency spectrum to eliminate the distortion signal present in said received input speech signal.