Computer system for speech recognition

• US 5,440,663 A
• Filed: 06/04/1993
• Issued: 08/08/1995
• Est. Priority Date: 09/28/1992
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Moese, Gerald, Bandara, Upali, Mohr, Karlheinz
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US08/072,624
• Time in Patent Office: 795 Days
• Field of Search: 395/2, 395/2.49, 395/2.6, 395/2.61, 395/2.64, 395/2.65, 381/43
• US Class Current: 704/255
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

The invention concerns a computer system for speech recognition with a word table and a triphone table in its memory. In response to a new word not contained in the word table and which is encountered by the computer system during recognition of an unknown spoken text, the fenemic baseform of this new word may be synthetically produced with the aid of a triphone table. For this purpose, the new word is decomposed into triphones, and the fenemic sequences associated with the individual triphones are read from the triphone table. The fenemic sequences are concatenated according to the phonetic baseform of the new word such that the fenemic baseform of the new word is obtained.

Claims

1. A computer system for speech recognition, comprising:

means for producing phonetic baseforms of words;

means for deriving triphones from the phonetic baseforms of the words;

means for producing all possible triphones of a language;

means for producing associated fenemic sequences from said all possible triphones;

a triphone table containing said all possible triphones and their associated fenemic sequences;

means for producing fenemic baseforms from the associated fenemic sequences;

a word table containing the words and the fenemic baseforms.

2. A computer system for speech recognition, comprising:

means for producing phonetic baseforms of words;

means for deriving triphones and uniphones from the phonetic baseforms of the words;

means for producing fenemic sequences associated with the triphones;

a triphone table containing the triphones and their associated fenemic sequences;

means for producing all possible uniphones of a language;

means for producing fenemic sequences associated with said all possible uniphones;

a uniphone table containing said all possible uniphones and their associated fenemic sequences;

means for producing fenemic baseforms from the fenemic sequences associated with the triphones and uniphones;

a word table containing the words and the fenemic baseforms.

3. The computer system of claim 2, further comprising:

means for deriving 1.5 phones from the phonetic baseforms of words;

means for producing all possible 1.5 phones of a language;

means for producing fenemic sequences associated with said all possible 1.5 phones;

a 1.5 phone table containing said all possible 1.5 phones and their associated fenemic sequences;

means for producing fenemic baseforms from the fenemic sequences associated with the 1.5 phones.

4. The computer system of claim 2, further comprising:

means for deriving classified triphones from the phonetic baseforms of words;

means for producing all possible classified triphones of a language;

means for producing fenemic sequences associated with said all possible classified triphones;

a classified triphone table containing said all possible classified triphones and their associated fenemic sequences;

means for producing fenemic baseforms from the fenemic sequences associated with the classified triphones.

5. A computer system for speech recognition, comprising:

means for producing phonetic baseforms of words;

means for deriving triphones and 1.5 phones from the phonetic baseforms of words;

means for producing fenemic sequences associated with the triphones;

a triphone table containing the triphones and their associated fenemic sequences;

means for producing all possible 1.5 phones of a language;

means for producing fenemic sequences associated with said all possible 1.5 phones;

a 1.5 phone table containing said all possible 1.5 phones and their associated fenemic sequences;

means for producing fenemic baseforms from the fenemic sequences associated with the triphones and 1.5 phones;

a word table containing the words and the fenemic baseforms.

6. A computer system for speech recognition comprising:

means for producing phonetic baseforms of words;

means for deriving triphones and classified triphones from the phonetic baseforms of words;

means for producing fenemic sequences associated with the triphones;

a triphone table containing the triphones and their associated fenemic sequences;

means for producing all possible classified triphones of a language;

means for producing fenemic sequences associated with said all possible classified triphones;

a classified triphone table containing said all possible classified triphones and their associated fenemic sequences;

means for producing fenemic baseforms from the fenemic sequences associated with the triphones and classified triphones;

a word table containing the words and the fenemic baseforms.

7. A method of producing fenemic sequences of a word, for use in a computer system for speech recognition, comprising the steps of:

(a) deriving a phonetic baseform of the word;

(b) decomposing the phonetic baseform of the word into triphones;

(c) producing fenemic sequences associated with the triphones;

(d) storing the triphones and the fenemic sequences associated with the triphones in a triphone table;

(e) iterating steps (a) through (d) until the triphone table contains all possible triphones of a language and the fenemic sequences associated with said all possible triphones.

8. The method of claim 7, further comprising the steps of:

(f) inputting a new word whose fenemic baseform is to be produced;

(g) deriving a phonetic baseform of the new word;

(h) decomposing the phonetic baseform of the new word into ordered triphones;

(i) selecting the fenemic sequences from the triphone table that correspond to the ordered triphones of step (h);

(j) concatenating the fenemic sequences of step (i) according to the ordered triphones of step (h);

(k) storing the fenemic sequences of step (j) in a word table as the fenemic baseform of the new word.

9. A method of producing fenemic sequences of a word, for use in a computer system for speech recognition, comprising the steps of:

(a) deriving a phonetic baseform of the word;

(b) decomposing the phonetic baseform of the word into uniphones;

(c) producing fenemic sequences associated with the uniphones;

(d) storing the uniphones and the fenemic sequences associated with the uniphones in a uniphone table;

(e) iterating steps (a) through (d) until the uniphone table contains all possible uniphones of a language and the fenemic sequences associated with said all possible uniphones.

10. The method of claim 9, further comprising the steps of:

(f) decomposing the phonetic baseform of step (a) into triphones;

(g) producing fenemic sequences associated with the triphones;

(h) storing the triphones and the fenemic sequences associated with the triphones in a triphone table;

(i) inputting a new word having a triphone not contained in the triphone table;

(j) deriving a phonetic baseform of the new word;

(k) decomposing the phonetic baseform of the new word into ordered triphones;

(l) selecting the fenemic sequences from the triphone table that correspond to the ordered triphones of step (k);

(m) selecting the fenemic sequence from the uniphone table that corresponds to the triphone of step (i);

(n) concatenating the fenemic sequences of step (l) and step (m) according to the ordered triphones of step (k);

(o) storing the fenemic sequence of step (n) in a word table as a fenemic baseform of the new word.

11. A method of producing fenemic sequences of a word, for use in a computer system for speech recognition, comprising the steps of:

(a) deriving a phonetic baseform of the word;

(b) decomposing the phonetic baseform into 1.5 phones;

(c) producing fenemic sequences associated with the 1.5 phones;

(d) storing the 1.5 phones and the fenemic sequences associated with the 1.5 phones in a 1.5 phone table;

(e) iterating steps (a) through (d) until the 1.5 phone table contains all possible 1.5 phones of a language and the fenemic sequences associated with said all possible 1.5 phones.

12. The method of claim 11, further comprising the steps of:

(f) decomposing the phonetic baseform of step (a) into triphones;

(g) producing fenemic sequences associated with the triphones;

(h) storing the triphones and the fenemic sequences associated with the triphones in a triphone table;

(i) inputting a new word having a triphone not contained in the triphone table;

(j) deriving a phonetic baseform of the new word;

(k) decomposing the phonetic baseform of the new word into ordered triphones;

(l) selecting the fenemic sequences from the triphone table that correspond to the ordered triphones of step (k);

(m) selecting the fenemic sequence from the 1.5 phone table that corresponds to the triphones of step (i);

(n) concatenating the fenemic sequences of step (l) and step (m) according to the ordered 1.5 phones of step (k);

(o) storing the fenemic sequence of step (n) in a word table as a fenemic baseform of the new word.

13. A method of producing fenemic sequences of a word, for use in a computer system for speech recognition, comprising the steps of:

(a) deriving a phonetic baseform of the word;

(b) decomposing the phonetic baseform into classified triphones;

(c) producing fenemic sequences associated with the classified triphones;

(d) storing the classified triphones and the fenemic sequences associated with the classified triphones in a classified triphone table;

(e) iterating steps (a) through (d) until the classified triphone table contains all possible classified triphones of a language and the fenemic sequences associated with said all possible classified triphones.

14. The method of claim 13, further comprising the steps of:

(f) decomposing the phonetic baseform of step (a) into triphones;

(g) producing fenemic sequences associated with the triphones;

(h) storing the triphones and the fenemic sequences associated with the triphones in a triphone table;

(i) inputting a new word having a triphone not contained in the triphone table;

(j) deriving a phonetic baseform of the new word;

(k) decomposing the phonetic baseform of the new word into ordered triphones;

(l) selecting the fenemic sequences from the triphone table that correspond to the ordered triphones of step (k);

(m) selecting the fenemic sequence from the classified triphone table that corresponds to the triphones of step (i);

(n) concatenating the fenemic sequences of step (l) and step (m) according to the ordered triphones of step (k);

(o) storing the fenemic sequences of step (n) in a word table as a fenemic baseform of the new word.