Phoneme recognition utilizing relative positions of reference phoneme patterns and input vectors in a feature space

• US 5,140,668 A
• Filed: 10/19/1990
• Issued: 08/18/1992
• Est. Priority Date: 11/10/1987
• Status: Expired due to Fees

• Current Assignee: NEC Corporation
• Original Assignee: NEC Corporation
• Inventors: Hattori, Hiroaki
• Primary Examiner(s): Shaw, Dale M.
• Assistant Examiner(s): Knepper, David D.
• Application Number:
  US07/600,049
• Time in Patent Office: 669 Days
• Field of Search: 364/513.5, 381/41-50, 395/2
• US Class Current: 704/254
• CPC Class Codes: G10L 15/10 using distance or distortion measures between unknown speech and reference templates
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

In a phoneme recognition apparatus, first distances between input vectors and reference vectors are determined. If the first distances are lower than a threshold value, the input vectors are identified as corresponding to the reference vectors. If the first distances are higher than the threshold value, the input vectors are identified as being indeterminate, and input differential vectors between the input vectors identified as corresponding to a reference vector and those identified as not corresponding to any of the reference vectors are determined. In addition, reference differential vectors between the reference vectors having corresponding input vectors and are those having no corresponding input vectors are determined. Second distances between the input differential vectors and the reference differential vectors are calculated and summed. The indeterminate input vectors are then identified as corresponding to the reference vectors in accordance with combined values of the first distances and the summed second distances.

Claims

1. A phoneme recognition apparatus comprising:

an input pattern memory for storing input vectors;

a reference pattern memory for storing reference vectors respectively representing phonemes;

first distance calculator means for determining first distances between said input vectors and said reference vectors;

a first distance memory for storing the first distances;

first discriminator means for determining whether the first distances are lower or higher than a prescribed threshold value and identifying a first portion of said input vectors which corresponds to those of the first distances which are determined as being lower than said threshold value as corresponding to said reference vectors and representing said first portion of said input vectors by first phoneme symbols, and identifying a second portion of said input vectors which corresponds to those of the first distances which are determined as being higher than said threshold value as being indeterminate and representing said second portion of said input vectors by second phoneme symbols;

a result memory for storing said first and second phoneme symbols;

input differential vector calculator means for determining input differential vectors between the input vectors identified by said first discriminator means as corresponding to said reference vectors and the input vectors identified by said first discriminator means as not corresponding to said reference vectors;

reference differential vector calculator means for determining reference differential vectors between the reference vectors having corresponding input vectors and the reference vectors having no corresponding input vectors;

second distance calculator means for determining second distances between said input differential vectors and said reference differential vectors;

adder means for providing summations of said second distances; and

second discriminator means responsive to a command signal applied thereto for identifying said second portion of said input vectors as corresponding to said reference vectors in accordance with combined values of the first distances stored in said first distance memory and said summed second distances and representing said second portion of said input vectors by third phoneme symbols.

2. A phoneme recognition apparatus as claimed in claim 1, further comprising:

control means for applying said command signal to said second discriminator means if said second phoneme symbols are stored in said result memory, and causing the first and third phoneme symbols to be read out of said result memory into an external circuit.

3. A phoneme recognition apparatus as claimed in claim 1, wherein said first discriminator means identifies one of said reference vectors which corresponds to a minimum of said first distances, determines whether said minimum of said first distances is lower or higher than said prescribed threshold value and identifies one of said input vectors which corresponds to the minimum of said first distances as corresponding to said identified reference vector if said minimum of said first distances is determined as being lower than said threshold value, and identifies one of said input vectors which corresponds to said minimum of said first distances as being indeterminate if said minimum of said first distances is determined as being higher than said threshold value, and wherein said second discriminator means identifies one of said reference vectors which corresponds to a minimum of said combined values and identifies one of said indeterminate input vectors which corresponds to said minimum combined value as corresponding to said reference vector identified as corresponding to said minimum combined value.

4. A method for identifying input vectors using reference vectors representing phonemes, comprising the steps of:

a) determining first distances between said input vectors and said references vectors;

b) determining whether the first distances are lower or higher than a prescribed threshold value and identifying a first portion of said input vectors which corresponds to those of the first distances determined as being lower than said threshold value as corresponding to said reference vectors, storing said first portion of said input vectors as first phoneme symbols into a result memory, identifying a second portion of said input vectors which corresponds to those of the first distances determined as being higher than said threshold value as being indeterminate, and storing said second portion of said input vectors as second phoneme symbols into said result memory;

c) determining input differential vectors between the input vectors identified by the step (b) as corresponding to said reference vectors and the input vectors identified by the step (b) as not corresponding to said reference vectors;

d) determining reference differential vectors between the reference vectors having corresponding input vectors and the reference vectors having no corresponding input vectors;

e) determining second distances between said input differential vectors and said reference differential vectors;

f) providing summations of said second distances; and

g) identifying said second portion of said input vectors as corresponding to said reference vectors in accordance with combined values of the first distances determined by the step (a) and the second distances summed by the step (f); and

storing the identified second portion of said input vectors into said result memory as third phoneme symbols.

5. A method as claimed in claim 4, wherein the step (b) comprises:

(b1) identifying one of said reference vectors as corresponding to a minimum of said first distances;

(b2) determining whether said minimum of said first distances is lower or higher than said prescribed threshold value;

(b3) if said minimum of said first distances is determined as being lower than said threshold value, identifying one of said input vectors which corresponds to the minimum of said first distances as corresponding to said reference vector identified by the step (b1); and

(b4) if said minimum of said first distances is determined as being higher than said threshold value, identifying one of said input vectors which corresponds to said minimum of said first distances as being indeterminate, and wherein the step (g) comprises;

(g1) identifying one of said reference vectors as corresponding to a minimum of said combined values; and

(g2) identifying one of said input vectors which corresponds to said minimum combined value as corresponding to the reference vector identified by the step (g1).