Voice-operated service

• US 5,384,833 A
• Filed: 05/05/1994
• Issued: 01/24/1995
• Est. Priority Date: 04/27/1988
• Status: Expired due to Term

• Current Assignee: Cisco Technology, Inc. (Cisco Systems, Inc.)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Cameron, Ian R.
• Primary Examiner(s): Kuntz, Curtis
• Assistant Examiner(s): Le, Huyen D.
• Application Number:
  US08/238,414
• Time in Patent Office: 264 Days
• Field of Search: 379/67, 379/88, 379/89, 381/41, 381/42, 381/43
• US Class Current: 704/275
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 17/22 Interactive procedures; Man-machine interfaces
• CPC Class Codes: H04M 2201/40 using speech recognition speech recognition per se G10L15/00
• CPC Class Codes: H04M 3/382 using authorisation codes or passwords
• CPC Class Codes: H04M 3/493 Interactive information services, e.g. directory enquiries ; Arrangements therefor, e.g. interactive voice response [IVR] systems or voice portals
• CPC Class Codes: Y10S 379/903 Password

Abstract

In a system for allocating a user identifier to a user of a voice-operated service, the user is requested (20) to utter words in a set of requested words. A speech recognizer (2) compares (24) each received word with a set of speech recognition templates, each template corresponding to the requested words. For each word uttered the speech recognizer generates (26) codes (W1, W2) which are the first and second best matched templates, and further generates (28) scores (D1, D2) indicating the extent of deviation of the received uttered words from the first and second best matched templates. The speech recognizer (2) then computes (32) a reliability criterion for each word by subtracting the scores (D1, D2) and comparing (34) the subtraction result with a threshold. Words having the greatest reliability criterion above the threshold are selected for the user identifier.

Claims

1. A method of allocating to a user of a voice operated service a user identifier comprising a plurality of words from a set of words, comprising:

(i) requesting the user to utter the words of the set;

(ii) comparing each uttered word with a set of speech recognition templates which correspond to the words of the set to obtain a measure of similarity for each uttered word with respect to each template of the set of speech recognition templates;

(iii) determining which of the words of the set meet a criterion related to a degree to which

(a) the measure of similarity of an uttered word to a template of said set of templates corresponding to a requested word exceeds

(b) the measure of similarity of the uttered word to another one of other templates of said set of templates with which the uttered word has the most similarity; and

(iv) selecting a sequence of words using only words of the set which meet the criterion.

2. A method according to claim 1 wherein the determination step (iii) involves determining whether a difference between the measure of similarity of subparagraph (a) and the measure of similarity of subparagraph (b) exceeds a threshold value.

3. A method according to claim 1 in which:

(a) each uttered word is compared with a first set of speech recognition templates corresponding to the words of the set and a second set of speech recognition templates corresponding to the words of the set;

(b) the measures of similarity obtained for the first set of templates is compared with the measure of similarity of the second set of templates to identify the set of templates which gives measures of similarity which are in general more indicative of similarity; and

(c) the determination as to which words meet the criterion is carried out using the identified set.

4. A method according to claim 3 in which two sets of templates correspond to male and female speech respectively.

5. A method according to claim 3 in which, in the selection of a sequence of words, a first of the words is taken from one of two subsets of the set of words, in dependence on which of the two sets of templates is identified as giving measures of similarity which are more indicative of similarity.

8. A method according to claim 1 in which:

(a) each uttered word is compared with a first set of speech recognition templates corresponding to the words of the set and a second set of speech recognition templates corresponding to the words of the set;

(b) the set of templates for which differences between

(i) measures of similarities of each uttered word to the corresponding requested word, and

(ii) measures of similarities of each uttered word to that one of the other templates of that set with which the uttered word has a measure of similarity most indicative of similarity,

are in general largest is identified; and

(c) the determination as to which words meet the criterion is carried out using the identified set.

9. A method according to claim 8 in which the two sets of templates correspond to male and female speech respectively.

10. A method according to claim 8 in which, in the selection of a sequence of words, the first of the words is taken from one of two subsets of the set of words, in dependence on which of the two sets of templates is identified as giving measures of similarity which are more indicative of similarity.

6. An apparatus for operating a voice-operated service comprising a speech recogniser operable to compare each word of a sequence of words supplied thereto with stored templates to obtain a measure of similarity, the templates comprising a first set containing first templates corresponding to each word of a set of words, and a second set containing second templates corresponding to each word of the said set of words, and control means arranged in operation so to control the recognizer that:

(a) the first word of the sequence is compared with the first templates corresponding to certain predetermined words of the set of words and the second templates corresponding to other words of the set of words;

(b) the measures of similarity are compared to determine which word of the set of words the said first word of the sequence most closely resembles; and

(c) subsequent words of the sequence are compared with either the first or the second templates, in dependence on which word of the set of words the said first word of the sequence has been determined as resembling.

7. An apparatus according to claim 6 in which the service is an interactive voice-operated service accessible via a telephone line.

11. An apparatus for allocating a user identifier to a user of a voice operated service, the user identifier comprising a plurality of words from a set of words, the apparatus comprising:

means for requesting the user to utter the words of the set and for obtaining uttered words corresponding to the words of the set;

speech recognition means for comparing each uttered word with a set of speech recognition templates which correspond to the words of the set to obtain a measure of similarity for each uttered word with respect to each template of the set of speech recognition templates;

control means for determining which of the words of the set meet a criterion related to a degree to which

(a) the measure of similarity of an uttered word to a template of said set of templates corresponding to a requested word exceeds

(b) the measure of similarity of the uttered word to another one of other templates of said set of templates with which the uttered word has the most similarity; and

and for selecting a sequence of words using only words of the set which meet the criterion.

12. The apparatus of claim 11 wherein the control means determines whether a difference between the measure of similarity of subparagraph (a) and the measure of similarity of subparagraph (b) exceeds a threshold value.

13. The apparatus of claim 11, wherein:

(a) the speech recognizer means compares each uttered word with a first set of speech recognition templates corresponding to the words of the set and to a second set of speech recognition templates corresponding to the words of the set;

(b) the control means compares the measures of similarity for the first set of templates with the measures of similarity of the second set of templates to identify the set of templates which gives measures of similarity which are in general more indicative of similarity; and

(c) the determination by the control means as to which words meet the criterion is carried out using the identified set.

14. The apparatus of claim 13, in which the first set of templates correspond to speech of a first gender and the second set of templates correspond to speech of a second gender.

15. The apparatus of claim 13, wherein, in the selection of a sequence of words by the control means, a first of the words is taken from one of two subsets of the set of words, in dependence on which of the two sets of templates is identified as giving measures of similarity which are more indicative of similarity.

16. The apparatus of claim 11, wherein

(a) the control means compares each uttered word with a first set of speech recognition templates corresponding to the words of the set and a second set of speech recognition templates corresponding to words of the set;

(b) the control means identifies the set of templates for which differences between

(i) measures of similarities of each uttered word to the corresponding requested word, and

(ii) measures of similarities of each uttered word to that one of the other templates of that set with which the uttered word has a measure of similarity most indicative of similarity,

are in general largest; and

(c) the determination by the control means as to which words meet the criterion is carried out using the identified set.

17. The apparatus of claim 16, in which the first set of templates correspond to speech of a first gender and the second set of templates correspond to speech of a second gender.

18. The apparatus of claim 16, wherein, in the selection of a sequence of words by the control means, a first of the words is taken from one of two subsets of the set of words, in dependence on which of the two sets of templates is identified as giving measures of similarity which are more indicative of similarity.