Speech training aid

• US 5,791,904 A
• Filed: 06/17/1996
• Issued: 08/11/1998
• Est. Priority Date: 11/04/1992
• Status: Expired due to Term

• Current Assignee: Aurix Ltd. (Avaya Incorporated)
• Original Assignee: The secretary of state for defence in her Britannic Majesty's Government of the United Kingdom of Great Britain And Northern Ireland
• Inventors: Russell, Martin J., Series, Robert W., Wallace, Julie L.
• Primary Examiner(s): Apley, Richard J.
• Assistant Examiner(s): Richman, Glenn E.
• Application Number:
  US08/664,661
• Time in Patent Office: 785 Days
• Field of Search: 434/156, 434/157, 434/167, 434/169, 434/178, 434/185
• US Class Current: 434/185
• CPC Class Codes: G09B 19/04 Speaking with audible presentation of the material to be studied G09B5/04
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/144 Training of HMMs
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/022 Demisyllables, biphones or triphones being the recognition units

Abstract

A speech training aid compares a trainee'"'"'s speech with models of speech, stored as sub-acoustic word models, and a general speech model to give an indication of whether or not the trainee has spoken correctly. An indication of how well the word has been pronounced may also be given. An adult operator enters the word to be tested into the training aid which then forms a model of that word from the stored sub-word speech models. The stored acoustic models are formed by first recording a plurality of words by a plurality of trainees from a given list of single words. These recordings are then processed off-line to give a basic acoustic model of an acceptable or correct sound for each phoneme in the context of the pre-and proceeding phonemes. The acoustic models are Hidden Markov Models. The limits of acceptable pronunciation, applied to different words and trainees, may be adjusted by variable penalty values applied in association with the general speech acoustic model. The training aid generates accumulated word costs for each trainee'"'"'s utterance and uses these costs to indicate correctness of pronunciation.

Claims

1. A speech training aid comprising;

an acoustic model store for storing acoustic models of speech sounds representing the processed results of speech by numerous person;

a dictionary for storing a list of words together with information to construct acoustic models of these words from the acoustic model store;

a network generator for generating an acoustic model of a required word together with an acoustic model of general speech sounds;

input means for requesting the generation of an acoustic model of the required word;

means for receiving a trainee'"'"'s utterances;

means for processing the trainee'"'"'s utterances into a form suitable for comparison with acoustic models of speech;

a speech pattern matcher for comparing the processed trainee'"'"'s utterances with the acoustic models of the required word and general speech; and

means for indicating whether or not the trainee'"'"'s utterances are the required word.

2. The training aid of claim 1 wherein the acoustic models are Hidden Markov Models.

3. The training aid of claim 1 wherein the trainee'"'"'s utterances are sampled and processed into an N-plurality of channels to provide an N-dimensional feature vector in successive time slot periods forming a sequence of feature vectors.

4. The training aid of claim 3 wherein the speech pattern matcher computes the relative probabilities of the sequence of feature vectors being the same as either the acoustic word model or the acoustic general speech model.

5. The training aid of claim 1 wherein the speech pattern matcher operates continually and the required word is recognised within continual speech.

6. The training aid of claim 1 wherein the speech pattern matcher computes the accumulated cost difference between the beginning and end of a recognised trainee-spoken word.

7. The training aid of claim 6 wherein the accumulated cost difference is used to give an indication of the accuracy of a trainee-spoken word.

8. The training aid of claim 1 and further including a variable penalty control arranged to cooperate with the acoustic model of general speech in forming an adjustable standard for the indication of accuracy of a trainee spoken word.

9. The training aid of claim 8 wherein the value of penalty is changed in accordance with set values associated with different required words.

10. The training aid of claim 9 wherein the set penalty values may be varied by an operator to suit individual trainees.

11. The speech training aid of claim 1 wherein the input means includes a keyboard for entering required words into the training aid for each testing session.

12. The speech training aid of claim 1 including means for prompting a trainee to speak the required word.

13. The speech training aid of claim 12 wherein the means for prompting includes a speech synthesizer and loudspeaker.

14. The speech training aid of claim 12 wherein the means for prompting includes a video monitor with screen for display of messages.

15. The speech training aid of claim 1 including means for recording the results of a session of a trainee'"'"'s speech, indicating words requested, words spoken correctly, and how correctly spoken.

16. The speech training aid of claim 1 wherein the acoustic word model is a series of sub-word models and general speech models, and the speech pattern matcher gives an indication of the correctness of utterance of the required word.

17. The speech training aid of claim 1 wherein the acoustic word model includes an additional acoustic model prior to the acoustic word model so that stuttering in a trainee'"'"'s utterances may be separately processed.

18. The training aid of claim 1 wherein the acoustic models of speech sounds includes acoustic models of phonemes.

19. The training aid of claim 1 wherein the acoustic models of speech sounds includes acoustic models of context sensitive phonemes.

20. The training aid of claim 1 wherein the network generator generates acoustic models of a plurality of words and the speech pattern matcher operates to recognize the required plurality of words within continual speech.

21. The training aid of claim 1 wherein the means for indicating includes means for indicating the accuracy of pronunciation of that required word.