Voice recognition process utilizing content addressable memory

• US 4,797,927 A
• Filed: 10/30/1985
• Issued: 01/10/1989
• Est. Priority Date: 10/30/1985
• Status: Expired due to Term

• Current Assignee: Grumman Aerospace Corporation So Oyster Bay Road Bethpage New York 11714
• Original Assignee: Grumman Aerospace Corporation (Northrop Grumman Corporation)
• Inventors: Schaire, Scott
• Primary Examiner(s): Salce, Patrick R.
• Assistant Examiner(s): Voeltz, Emanuel T.
• Application Number:
  US06/792,994
• Time in Patent Office: 1,168 Days
• Field of Search: 382/10, 382/23, 382/25, 382/29, 381/41-50, 364/513, 364/513.5
• US Class Current: 704/252
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence

Abstract

A voice recognition system digitizes sequentially spoken words forming a vocabulary. As each word is spoken during an initializing phase, its digital value forms a score which corresponds directly to an address in memory. During a later voice recognition phase, any word spoken is scored and the word is read from an indirectly addressed memory without the necessity of comparisons with the data stored in each memory location.

Claims

1. A voice recognition method comprising:

(a) an initializing phase including the steps;

converting a preset vocabulary list of spoken words to digitized form;

performing time compression of the digitized words to standardize the length of each word wherein the compressed word has a corresponding digital score;

formatting a main memory with addresses directly corresponding to each score, the data at each address being the position of the respective word in the vocabulary; and

(b) wherein the recognition phase includes the steps;

converting a spoken word to a digitized form;

performing time compression of the digitized word to standardize the length of each word wherein each compressed word has a corresponding digital score;

addressing the main memory at the location directly corresponding to the score;

reading out, from main memory, the position of the word in the vocabulary;

addressing a vocabulary memory at a memory location corresponding to the position of the word in the vocabulary; and

reading out a word stored at the vocabulary memory location, corresponding to the word spoken during the recognition phase.

2. The method set forth in claim 1 together with the initializing phase step of speaking the words of the vocabulary more than once to establish separate scores for variations of the same word.

3. The method set forth in claim 2 wherein additional addresses of the main memory directly correspond to the scores of the variations, the data at each additional address corresponding to the position of a respective word in the vocabulary.

4. The method set forth in claim 2 wherein a previously unscored word occurring after time compression in the recognition phase undergoes recognition during the steps including:

performing a bidirectional linear search of the main memory to find the word having the closest score to that of the unscored word;

querying a user as to whether the unscored word has been correctly searched and, if not;

identifying the vocabulary position of the unscored word; and

storing the vocabulary position of the word at the score address.

5. In a voice recognition method wherein a main memory has addresses directly corresponding to digital values of spoken words, the contents at each address corresponding to the position of the word in a vocabulary, recognition including the steps:

converting a spoken word to a digitized form;

performing time compression of the digitized word to standardize the length of each word wherein each compressed word has a corresponding digital score;

addressing the main memory at the location directly corresponding to the score;

reading out, from main memory, the position of the word in the vocabulary;

addressing a vocabulary memory at a memory location corresponding to the position of the word in vocabulary; and

reading out a word stored at the vocabulary memory location, corresponding to the word spoken during the recognition phase.

6. The method set forth in claim 5 wherein a previously unscored word occurring after time compression in the recognition phase undergoes recognition during the steps including:

performing a bidirectional linear search of the main memory to find the word having the closest score to that of the unscored word;

querying a user as to whether the unscored word has been correctly searched and, if not;

identifyIng the vocabulary position of the unscored word; and

storing the vocabulary position of the word at the score address.

7. A voice recognition system comprising:

an analog-to-digital converting means for digitizing spoken words during initializing and recognition phases;

means for time compressing the digital words during the initializing and recognition phases thereby representing digital scores for the words;

main memory means connected to the output of the time-compressing means for storing the vocabulary position of each word spoken during an initializing phase at the address directly corresponding to the score of the word; and

vocabulary memory means storing the vocabulary words at addresses corresponding to the position of the word in the vocabulary;

wherein the score of each word spoken during a recognition phase directly addresses the main memory to derive the vocabulary position of the spoken word thus enabling the vocabulary memory to be indirectly addressed for reading out the recognized word.

8. The structure set forth in claim 7 together with a communicating port connected between the main and vocabulary memories to allow indirect addressing of the vocabulary memory during a recognition phase for retrieval of each recognized word.

9. The structure set forth in claim 7 together with terminal means connected in circuit to the memories for allowing entry in the system of the vocabulary position of a word unrecognized during a recognition phase;

and

adding the score of the unrecognized word to the main memory to permit its future recognition.

10. The structure set forth in claim 9 together with a communicating port connected between the main and vocabulary memories to allow indirect addressing of the vocabulary memory during a recognition phase for retrieval of each recognized word.