Speech recognition device

• US 5,058,167 A
• Filed: 05/02/1990
• Issued: 10/15/1991
• Est. Priority Date: 07/16/1987
• Status: Expired due to Term

• Current Assignee: Fujitsu Limited
• Original Assignee: Fujitsu Limited
• Inventors: Kimura, Shinta
• Primary Examiner(s): Harkcom, Gary V.
• Application Number:
  US07/520,853
• Time in Patent Office: 531 Days
• Field of Search: 381/41-46, 364/513.5
• US Class Current: 704/249
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence

Abstract

A speech recognition device for recognizing a speech by analyzing an input speech characteristic portion extracted from the input speech and comparing the extracted characteristic portion with speech dictionary templates, the device including a speech holding unit for receiving and storing the received input speech, a speech division detection unit connected with the speech holding unit for detecting divisions of the speech read from the speech holding unit, and an instruction signal supply unit for supplying an instruction for recognizing the speech read from the speech holding unit, wherein the instruction signal supply unit is operated after an input of a speech to be recognized, and based on the operation of the instruction signal supply unit, a recognition processing of only a correct input speech is carried out.

Claims

1. A speech recognition device for recognizing only correct input speech from an input speech waveform by extracting a speech characteristic portion from the input speech waveform of input speech and comparing the speech characteristic portion with speech templates, said device comprising:

speech holding means for receiving and storing input acoustic signals corresponding to the input speech;

instruction signal supply means for supplying a nearest past speech instruction signal after an input of speech to be recognized, the nearest past speech instruction signal is input from a user of said device;

speech division detection means, connected to said speech holding means and said instruction signal supply means, for receiving the input acoustic signals corresponding to the input speech and for detecting divisions of the input speech read from said speech holding means to determine a nearest past speech stored in said speech holding means based upon the nearest past instruction signal and for extracting the speech characteristic portion corresponding to the nearest past speech; and

recognition means for comparing the speech characteristic portion corresponding to the nearest past speech with the speech templates to produce a recognized result.

2. A device according to claim 1,

wherein the recognized result has at least one homophone, and

wherein said device further comprises:

possible-choice and homophone selection means for providing alternative recognized results based on the at least one homophone of the recognized result when the recognized result is not the homophone to be recognized; and

possible-choice selection signal supply means for supplying a possible-choice selection signal to said possible-choice and homophone selection means to select one of the alternative recognized results.

3. A device according to claim 1,

wherein the recognized result includes at least one homophone,

wherein said device further comprises possible-choice and homophone selection means for providing alternative recognized results based on the at least one homophone of the recognized result, and

wherein the nearest past speech instruction signal selects one of the alternative recognized results.

4. A device according to claim 2, wherein said speech holding means includes means for interrupting the receiving of the input speech at said speech holding means when a selection operation by said possible-choice selection signal supply means is carried out.

5. A device according to claim 1, wherein said speech division detection means further comprises island waveform detecting means for detecting islands in the input speech waveform to enable detection of the divisions of the input speech.

6. A device according to claim 5, wherein the input speech waveform is inspected in said island waveform detecting means by tracing the input speech waveform in reverse from an occurrence of the nearest past speech instruction signal from said instruction signal supply means by using predetermined threshold values to determine existence of island waveforms.

7. A device according to claim 1, wherein said device further comprises:

dictionary means for storing homophones for a speech pattern;

possible-choice and homophone selection means for providing alternative recognized results based on the homophones stored in said dictionary means corresponding to the recognized result, whereby a plurality of labels for the homophones are output one after another in accordance with a predetermined priority; and

possible-choice selection signal supply means for supplying a possible-choice selection signal to said possible-choice and homophone selection means to select one of the alternative recognized results.