Speech recognition system using shared speech models for multiple recognition processes

• US 6,061,653 A
• Filed: 07/14/1998
• Issued: 05/09/2000
• Est. Priority Date: 07/14/1998
• Status: Expired due to Term

• Current Assignee: WSOU Investments, LLC (WSOU Holdings, LLC)
• Original Assignee: Alcatel USA Sourcing L.P. (Nokia Corporation)
• Inventors: Spiess, Jeffrey J., Mowry, Dearborn R., Fisher, Thomas D.
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Storm, Donald L.
• Application Number:
  US09/115,782
• Time in Patent Office: 665 Days
• Field of Search: 704/256, 704/251, 704/252, 704/254, 704/255, 704/243, 704/244, 379/88.01-88.04
• US Class Current: 704/256.5
• CPC Class Codes: G10L 15/142 Hidden Markov Models [HMMs]
• CPC Class Codes: G10L 15/285 Memory allocation or algorithm optimisation to reduce hardware requirements
• CPC Class Codes: G10L 15/32 Multiple recognisers used in sequence or in parallel; Score combination systems therefor, e.g. voting systems

Abstract

A method of operating a speech recognition system. The method loads a speech model from a storage facility into a memory accessible by a processor. This loading step includes two steps. A first of these steps loads process-independent state data representative of a plurality of states of the speech model. A second of these steps loads process-specific state data representative of the plurality of states of the speech model. The speech recognition system then performs a first speech recognition process with the processor by accessing the process-independent state data and a first portion of the process-specific state data. The speech recognition system also performs a second speech recognition process with the processor, where the second process also accesses the process-independent state data but further accesses a second portion of the process-specific state data different than the first portion of the process-specific state data. The speech model data may be the states, transitions, and scores of a Hidden Markov Model or pointers to the model may be loaded. In response to a third recognition process, loading a third portion of process-specific data invalidates the first portion.

Claims

1. A method of operating a speech recognition system, comprising the steps of:

loading a speech model from a storage facility into a memory accessible by a processor, comprising the steps of;

loading process-independent state data representative of a plurality of states of the speech model;

loading process-specific state data representative of the plurality of states of the speech model;

performing a first speech recognition process with the processor by accessing the process-independent state data and a first portion of the process-specific state data; and

performing a second speech recognition process with the processor by accessing the process-independent state data and a second portion of the process-specific state data different than the first portion of the process-specific state data.

2. The method of claim 1:

wherein the step of loading the speech model further comprises;

loading process-independent transition data representative of a plurality of transitions of the speech model;

loading process-specific transition data representative of the plurality of transitions of the speech model;

and further comprising the steps of;

performing the first speech recognition process with the processor by accessing the process-independent transition data and a first portion of the process-specific transition data; and

performing a second speech recognition process with the processor by accessing the process-independent transition data and a second portion of the process-specific transition data different than the first portion of the process-specific transition data.

3. The method of claim 2:

wherein the step of loading the speech model further comprises the step of invalidating the first portion of the process-specific transition data responsive to loading a third portion of process-specific transition data into the memory;

and further comprising the step of performing a third speech recognition process with the processor by accessing the process-independent transition data and the third portion of the process-specific transition data.

4. The method of claim 2 wherein the process-independent transition data comprises data selected from the group consisting of a pointer to a destination state, a transition score, and a pointer to an observation model.

5. The method of claim 2 wherein the process-specific transition data comprises data selected from the group consisting of an active score and a pointer to an active score.

6. The method of claim 2:

wherein the step of loading the speech model further comprises the steps of;

invalidating the first portion of the process-specific transition data responsive to loading a third portion of process-specific transition data into the memory;

invalidating the first portion of the process-specific state data responsive to loading a third portion of process-specific state data into the memory;

and further comprising the step of performing a third speech recognition process with the processor by accessing the process-independent state data and the third portion of the process-specific state data and the third portion of the process-specific transition data.

7. The method of claim 1:

wherein the step of loading the speech model further comprises the step of invalidating the first portion of the process-specific state data responsive to loading a third portion of process-specific state data into the memory;

and further comprising the step of performing a third speech recognition process with the processor by accessing the process-independent state data and the third portion of the process-specific state data.

8. The method of claim 1 wherein the step of loading a speech model comprises loading a hidden Markov model.

9. The method of claim 1 wherein the step of loading a speech model comprises loading a phonetic speech model.

10. The method of claim 1 wherein the process-independent state data comprises data selected from the group consisting of a pointer to a host model, a number of transitions for a given state, and a state stop score.

11. The method of claim 1 wherein the process-specific state data comprises data selected from the group consisting of an active score, a pointer to an active score, and a cumulative number of occurrences of a state.

12. The method of claim 1 wherein the step of loading a speech model further comprises loading model data, the model data comprising data selected from the group consisting of a total size of the speech model, an identifier for the speech model, and number of states in the speech model.

13. The method of claim 1 wherein the speech recognition system is a telephony speech recognition system.

14. A method of operating a speech recognition system, comprising the steps of:

loading a speech model from a storage facility into a memory accessible by a processor, comprising the steps of;

loading process-independent data representative of a group selected from one of a plurality of states of the speech model and a plurality of transitions of the speech model;

loading process-specific data representative of the group selected from one of a plurality of states of the speech model and a plurality of transitions of the speech model;

performing a first speech recognition process with the processor by accessing the process-independent data and a first portion of the process-specific data; and

performing a second speech recognition process with the processor by accessing the process-independent data and a second portion of the process-specific data different than the first portion of the process-specific data.

15. A speech recognition system, comprising:

a storage facility for storing a speech model;

a plurality of processors; and

a computer programmed to perform the steps of;

loading the speech model from the storage facility into a memory accessible by one of the plurality of processors, comprising the steps of;

loading process-independent data representative of a group selected from one of a plurality of states of the speech model and a plurality of transitions of the speech model;

loading process-specific data representative of the group selected from one of a plurality of states of the speech model and a plurality of transitions of the speech model;

wherein one of the plurality of processors is programmed to perform the steps of;

performing a first speech recognition process with the processor by accessing the process-independent data and a first portion of the process-specific data; and

performing a second speech recognition process with the processor by accessing the process-independent data and a second portion of the process-specific data different than the first portion of the process-specific data.

16. The system of claim 15 wherein the one of the plurality of processors is programmed to further perform the steps of:

invalidating the first portion of the process-specific data responsive to loading a third portion of process-specific data into the memory, the third portion representative of the group selected from one of a plurality of states of the speech model and a plurality of transitions of the speech model; and

performing a third speech recognition process with the processor by accessing the process-independent data and the third portion of the process-specific data.

17. The system of claim 15 wherein the process-independent data represents the plurality of transitions of the speech model and comprises data selected from the group consisting of a pointer to a destination state, a transition score, and a pointer to an observation model.

18. The system of claim 15 wherein the process-specific data represents the plurality of transitions of the speech model and comprises data selected from the group consisting of an active score and a pointer to an active score.

19. The system of claim 15 wherein the process-independent data represents the plurality of states of the speech model and comprises data selected from the group consisting of a pointer to a host model, a number of transitions for a given state, and a state stop score.

20. The system of claim 15 wherein the process-specific data represents the plurality of states of the speech model and comprises data selected from the group consisting of an active score, a pointer to an active score, and a cumulative number of occurrences of a state.

21. The system of claim 15 wherein the a speech model a hidden Markov model.

22. The system of claim 15 wherein the speech model comprises a phonetic speech model.