Apparatus for automatic speech recognition

• US 4,624,008 A
• Filed: 03/09/1983
• Issued: 11/18/1986
• Est. Priority Date: 03/09/1983
• Status: Expired due to Term

• Current Assignee: ITT Corporation (ITT, Inc.)
• Original Assignee: International Telephone and Telegraph Corporation
• Inventors: Vensko, George, Carlin, Lawrence, Nelson, Mark W., Smith, A. Richard
• Primary Examiner(s): Kemeny, E. S. Matt
• Assistant Examiner(s): Salotto, John J.
• Application Number:
  US06/473,422
• Time in Patent Office: 1,350 Days
• Field of Search: 381/41, 381/42, 381/43, 381/44, 381/45, 381/46, 381/47, 381/48, 381/49, 381/50, 364/513, 364/513.5
• US Class Current: 704/253
• CPC Class Codes: G10L 15/00 Speech recognition G10L17/00 takes precedence
• CPC Class Codes: G10L 15/12 using dynamic programming techniques, e.g. dynamic time warping [DTW]

Abstract

An apparatus and method for recognition of sentences comprised of utterances separated by short pauses, the utterances representative of both isolated words and connected words. Speech to be recognized is converted into frames of digital signals. Selected ones of the frames of digital signals are compared with isolated word and connected word templates stored in a template memory. Recognition of isolated words is done by comparing the selected frames of digital signals with the isolated word templates in accordance with a windowed dynamic programming algorithm having path boundary control, while connected word recognition is accomplished by comparing selected frames of digital signals with the connected word templates in accordance with a full DPA having path score normalization and utterance frame penalty calculation capability. Variable frame rate encoding is used to identify the selected frames of digital signals. Syntax control includes selecting the isolated word and connected word templates to be compared after each utterance and includes combining the recognized isolated words and connected words into sentences in accordance with the predefined syntax after the end of the sentence has been detected. A logical set of connected words are the connected digits.

Claims

1. An apparatus for recognizing sentences of speech, said sentences comprised of utterances separated by pauses and including both isolated words and connected words taken from a predefined finite size vocabulary, said apparatus comprising:

means including a bandpass filter bank for forming frames of digital signals representative of said sentences of speech; and

microprocessor means for recognizing said sentences from said frames of digital signals in accordance with a predefined syntax;

wherein said means for recognizing said sentences further comprises;

memory means for storing said predefined vocabulary as isolated word templates and partial phrase connected word templates;

means for detecting the endpoints of said utterances and for generating an output representative of such endpoint detection;

isolated word recognizing means responsive to the output of said endpoint detection means for recognizing said isolated words by comparing selected ones of said isolated word templates with said frames of digital signals;

connected word recognizing means responsive to the output of said endpoint detection means for recognizing said connected words by comparing selected ones of said partial phrase connected word templates with said frames of digital signals; and

means for selecting said selected ones of said isolated word and partial phrase connected word templates and combining said recognized isolated words and connected words into said sentences in accordance with said predefined syntax.

2. The invention of claim 1, wherein said means for recognizing said connected words comprises recognition of connected digits.

3. The invention of claim 1, wherein said comparison of said selected isolated word templates and said frames of digital signals is done in accordance with a windowed dynamic programming algorithm (DPA).

4. The invention of claim 3, wherein said comparison of said selected isolated word templates and said frames of digital signals is done in accordance with a windowed DPA having path boundary control.

5. The invention of claim 1, wherein said comprarison of said selected partial phrase connected word templates and said frames of digital signals is done in accordance with a full dynamic programming algorithm (DPA).