Data processing apparatus and method for use in speech recognition

• US 4,567,606 A
• Filed: 11/03/1982
• Issued: 01/28/1986
• Est. Priority Date: 11/03/1982
• Status: Expired due to Term

• Current Assignee: ITT Corporation (ITT, Inc.)
• Original Assignee: International Telephone and Telegraph Corporation
• Inventors: Vensko, George, Potter, John, Smith, Allen R., Carlin, Lawrence
• Primary Examiner(s): KEMENY, EMANUEL
• Application Number:
  US06/439,018
• Time in Patent Office: 1,182 Days
• Field of Search: 381/41-45, 364/513.5
• US Class Current: 704/251
• CPC Class Codes: G06F 15/167 using a common memory, e.g. mailbox
• CPC Class Codes: G06F 3/16 Sound input; Sound output speech processing G10L
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements

Abstract

A continuous speech recognition system includes a plurality of processors doing template comparisons of speech data. Each processor has an associated memory shared with the other processors by direct memory access (DMA) through a shared data bus. The DMA circuitry is distributed between the processors to eliminate redundancy, since if each processor had a full DMA circuit, one of the circuits would be idle when the processors communicated.

Claims

1. A distributed processing apparatus for performing a continuous speech recognition function comprising:

a plurality of processing means coupled by a remote bus and each having a shared memory associated therewith and each for performing local processing on data stored in said associated shared memory; and

data transfer means associated with each of said processing means and with said associated shared memory and coupled thereto for transferring at least a portion of said data between said shared memories across said remote bus;

direct memory access circuit means (DMA) for preventing interference with local processing of the remaining portions of said data by said plurality of processing means during said transfer of data.

2. The data processing apparatus of claim 1, wherein said data transfer means further comprises:

a shared data bus associated with each of said shared memories and processing means and coupled thereto for transferring data between said shared memories and said remote bus and between said shared memories and said processing means associated therewith;

a remote bus controller for controlling the transfer of data across said remote bus;

a plurality of circuit means for effecting data transfer across said shared data bus including;

DMA circuitry for DMA transfer of said data between said shared memories in response to input from at least one of said processing means prior to said DMA transfer;

programmable I/O circuitry for programmed transfer of data across said remote bus including providing said input to said DMA circuitry; and

shared controller circuitry for controlling the data transfer across said shared data bus.

3. The data processing apparatus of claim 2, wherein one of said plurality of processing means is a master processing means and the other processing means are slave processing means, said master processing means for controlling said data transfer including providing input to DMA circuitry.

4. The data processing apparatus of claim 3, wherein said shared controller circuitry comprises:

decode circuitry for decoding local addresses from said associated processing means to generate local and DMA control signals and for decoding remote addresses from said master processing means to generate remote and DMA control signals;

circuitry for generating prioritized local, remote and DMA requests for said shared bus in response to a first portion of said control signals;

circuitry for synchronizing said prioritized request signals with the local processing of shared memory data by said slave processor whereby said local processing is uninterrupted; and

clocking and enabling circuitry for transferring data between said shared data bus and a selected one of said plurality of circuit means.

5. An apparatus for recognizing speech continuously spoken comprising:

a plurality of processing means coupled together by a remote bus and each having a shared memory associated therewith, said processing means including means for converting said continuously spoken speech into parametric data, means for comparing said parametric data with stored template data, and means for analyzing the results of said comparison; and

data transfer means associated with each of said processing means and associated shared memory and coupled thereto for transferring said parametric data, template data, and results data among said shared memories; and

direct memory access means for preventing interference with the converting, comparing or analyzing of remaining portions of said data during said transferring.

6. The apparatus of claim 5, wherein said plurality of processing means comprises:

a front end processing means including a speech converting means for sampling said continuously spoken words a plurality of times and converting said sampled words into digital data, said front end processing means for converting said digital data into frames of parametric data corresponding to said samples;

at least two template processing means for storing said template data and for comparing in accordance with a predetermined algorithm said frames of parametric data with said template data to provide results data; and

master processing means for controlling the transfer of said frames of parametric data to said at least two template processing means, the transfer of template data among said template processing means and said results data to said master processing means with each new frame of parametric data.

7. The apparatus of claim 6, wherein said plurality of template processing means comprises at least four template processing means and said apparatus has a vocabulary of at least 300 words.

8. A method of automatically recognizing continuously spoken words in real time using a speech processor which converts said continuously spoken words to parametric data and using at least two template processors which store templates of a vocabulary of words that can be recognized and which compare said parametric data with said templates, said method comprising the step of:

transferring said parametric data and said templates among said speech processor and said at least two template processors without interfering with the conversion of said continuously spoken words to said parametric data or with the comparison of said parametric data with said templates in the course of said transferring whereby the automatic recognition of continuously spoken words in real time is expedited.

9. A method of recognizing continuously spoken words comprising:

converting said continuously spoken words into frames of parametric data more suitable for speech processing in a front end data processor;

storing a vocabulary of words to be recognized as templates in at least two template processors;

transferring said frames of parametric data to said at least two template processors for comparison of said frames with said stored templates; and

redistributing said stored templates among said at least two template processors for comparison with a new frame of parametric data while a previous frame of parametric data is being processed.

10. The method of claim 9, wherein the step of transferring said frames of parametric data and the step of redistributing said stored templates is done by a plurality of means for direct memory access.