Speaker-dependent connected speech word recognizer

• US 4,831,551 A
• Filed: 10/13/1987
• Issued: 05/16/1989
• Est. Priority Date: 01/28/1983
• Status: Expired due to Term

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Schalk, Thomas B., Doddington, George R.
• Primary Examiner(s): KEMENY, EMANUEL
• Application Number:
  US07/110,062
• Time in Patent Office: 581 Days
• Field of Search: 381/41-50, 364/513.5
• US Class Current: 704/233
• CPC Class Codes: G10L 15/12 using dynamic programming techniques, e.g. dynamic time warping [DTW]

Abstract

Speech recognition is improved using reference pattern templates which have an added noise signal (noise floor) to avoid LPC high-gain synthesizer instability at low signal levels. Also, input signal frames have a length one-half that of reference frames whereby dynamic time warp computation steps are cut almost in half.

Claims

1. A method for speech recognition, comprising the steps of:

storing reference templates containing LPC enters corresponding to a plurality of portions of words in a predefined vocabulary, wherein said templates further include in the LPC parameters a noise signal having a magnitude which is a preselected fraction of the magnitude of the portions of words;

receiving speech inputs, and transforming the inputs into a plurality of frames of LPC parameters; and

calculating error values indicating the distance between each input frame and each of the reference templets, wherein an utterance is hypothesized to be that of the set of reference templates having the lowest error values.

2. A method for speech recognition, comprising the steps of:

storing reference templates containing LPC parameters corresponding to a plurality of portions of words in a predefined vocabulary, each of the reference templates having a frame length;

receiving speech inputs, and transforming the inputs into a plurality of frames of LPC parameters, the input frames having a frame length less than that of the reference frames; and

calculating error values indicating the distance between each input frame and each of the reference templates using dynamic time warping, wherein an utterance is hypothesized to be that of the set of reference templates having the lowest error values.

3. The method of claim 2, wherein the input frames have a frame length which is one-half that of the reference frames.

4. A method for recognizing speech, comprising the steps of:

storing reference templates containing LPC parameters corresponding to a plurality of portions of words in a predefined vocabulary, wherein the LPC parameters include a noise signal having an energy which is proportional to the energy of the word portions stored in said templates;

receiving speech inputs, and transforming the inputs into a plurality of frames of LPC parameters; and

calculating error values indicating the distance between each input frame and each of the reference templets, wherein an utterance is hypothesized to be that of the set of reference templates having the lowest error value.