Method and apparatus for improved speech recognition by modifying a pronunciation dictionary based on pattern definitions of alternate word pronunciations

• US 6,389,394 B1
• Filed: 02/09/2000
• Issued: 05/14/2002
• Est. Priority Date: 02/09/2000
• Status: Expired due to Term

• Current Assignee: SpeechWorks International, Inc. (Microsoft Corporation)
• Original Assignee: SpeechWorks International, Inc. (Microsoft Corporation)
• Inventors: Fanty, Mark
• Primary Examiner(s): Dorvil, Richemond
• Application Number:
  US09/501,341
• Time in Patent Office: 825 Days
• Field of Search: 704/260, 704/254, 704/256, 704/244, 704/200, 704/251, 704/255, 704/257, 704/258, 704/266, 704/249, 704/246
• US Class Current: 704/249
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

An approach for automatically modifying a pronunciation dictionary in a speech recognition system based on patterns of alternate pronunciations is described. A representation of the pronunciation dictionary, such as a plurality of dynamically linked phoneme values, is obtained. One or more pattern definitions are obtained. The pattern definitions specify zero or more phonemes to be substituted for zero or more phonemes of all words in the pronunciation dictionary. The linked phoneme values are modified by adding, for each path of each word, alternate paths that use each of the substitute phonemes according to the pattern definitions, thereby creating an expanded set of dynamically linked phoneme values. One or more example pronunciations of a particular word are then obtained. One or more best paths through the expanded set of phoneme values are determined for each of the example pronunciations and used to find the overall best path(s). For the overall best path(s), an alternate word pronunciation is constructed by converting each path into a pronunciation using the format of the pronunciation dictionary. The pronunciation dictionary is modified by adding each alternate word pronunciation. As a result, a modified pronunciation dictionary is created that accounts for alternate pronunciations as actually spoken by users of a particular speech recognition application.

Claims

1. A method for modifying a pronunciation dictionary in a speech recognition system to include one or more alternate pronunciations, wherein each pronunciation in the pronunciation dictionary is represented as one or more dynamically linked phoneme values, the method comprising the computer-implemented steps of:

receiving one or more phoneme substitution patterns that define one or more substitute phoneme values for one or more of the dynamically linked phoneme values;

adding the substitute phoneme values and new links therefor to the dynamically linked phoneme values;

determining one or more best paths among the dynamically linked phoneme values for a particular word under consideration; and

modifying the pronunciation dictionary by adding a pronunciation that represents each of the best paths.

2. A method as recited in claim 1, wherein receiving one or more phoneme substitution patterns comprises the step of receiving one or more phoneme substitution patterns each comprising (a) information identifying a first phoneme and (b) one or more phonemes that are to be substituted for the first phoneme during recognition of a word containing the first phoneme.

3. A method as recited in claim 1, wherein receiving one or more phoneme substitution pattern comprises the step of receiving at least one phoneme substitution pattern that comprises (a) information identifying a first phoneme and (b) information indicating that the first phoneme is to be deleted from the dynamically linked phoneme values.

4. A method as recited in claim 1, wherein determining one or more best paths further comprises the steps of receiving two or more example pronunciations for a particular word under consideration and determining one or more best paths among the dynamically linked phoneme values for each of the example pronunciations.

5. A method as recited in claim 1, wherein determining one or more best paths further comprises the steps of receiving two or more example pronunciations for a particular word under consideration based on measurement data that is generated and stored by the speech recognition system during its execution.

6. A method as recited in claim 1, wherein determining one or more best paths further comprises the steps of converting each of the best paths into pronunciations according to a format of representing pronunciations that is used in the pronunciation dictionary.

7. A method for modifying a pronunciation dictionary in a speech recognition system to include one or more alternate pronunciations, wherein each pronunciation in the pronunciation dictionary is represented during use of the speech recognition system as one or more paths among nodes in a dynamically linked phoneme network, wherein phonemes in each pronunciation are represented by arcs between nodes in the network, the method comprising the computer-implemented steps of:

receiving one or more phoneme substitution patterns that define one or more substitute values for one or more of the phonemes in the network;

adding the substitute values and new arcs therefor to the network;

determining one or more best paths through the network for a particular word under consideration; and

modifying the pronunciation dictionary by adding a pronunciation that represents each of the best paths.

8. A method as recited in claim 7, wherein the step of receiving one or more phoneme substitution patterns comprises the step of receiving one or more phoneme substitution patterns each comprising (a) information identifying a first phoneme and (b) one or more phonemes that are to be substituted for the first phoneme in the network.

9. A method as recited in claim 7, wherein the step of receiving one or more phoneme substitution patterns comprises receiving at least one phoneme substitution pattern that comprises (a) information identifying a first phoneme and (b) information indicating that the first phoneme is to be deleted from the network.

10. A method as recited in claim 7, wherein determining one or more best paths further comprises the steps of receiving two or more example pronunciations for a particular word under consideration and determining one or more best paths through the network for each of the example pronunciations.

11. A method as recited in claim 7, wherein determining one or more best paths further comprises the steps of receiving two or more example pronunciations for a particular word under consideration based on measurement data that is generated and stored by the speech recognition system during its execution.

12. A method as recited in claim 7, wherein determining one or more best paths further comprises the steps of converting each of the best paths into pronunciations according to a format of representing pronunciations that is used in the pronunciation dictionary.

13. A method for modifying a pronunciation dictionary in a speech recognition system to include one or more alternate pronunciations, wherein each pronunciation in the pronunciation dictionary is represented during use of the speech recognition system as one or more paths among nodes in a dynamically linked phoneme network, wherein phonemes in each pronunciation are represented by arcs between nodes in the network, the method comprising the computer-implemented steps of:

receiving one or more phoneme substitution patterns that define one or more substitute values for one or more of the phonemes in the network;

adding the substitute values and new arcs therefor to the network;

receiving a plurality of example pronunciations for a particular word under consideration;

determining one or more best paths through the network for the word under consideration by carrying out recognition of the example pronunciations using the network; and

modifying the pronunciation dictionary by adding a pronunciation that represents each of the best paths.

14. A computer-readable medium carrying one or more sequences of one or more instructions for modifying a pronunciation dictionary in a speech recognition system to include one or more alternate pronunciations, wherein each pronunciation in the pronunciation dictionary is represented as one or more dynamically linked phoneme values, the one or more sequences of one or more instructions including instructions which, when executed by one or more processors, cause the one or more processors to perform the steps of:

receiving one or more phoneme substitution patterns that define one or more substitute phoneme values for one or more of the dynamically linked phoneme values;

adding the substitute phoneme values and new links therefor to the dynamically linked phoneme values;

determining one or more best paths among the dynamically linked phoneme values for a particular word under consideration; and

modifying the pronunciation dictionary by adding a pronunciation that represents each of the best paths.

15. A computer-readable medium as recited in claim 14, wherein instructions for receiving one or more phoneme substitution patterns further comprise instructions for receiving one or more phoneme substitution patterns each comprising (a) information identifying a first phoneme and (b) one or more phonemes that are to be substituted for the first phoneme during recognition of a word containing the first phoneme.

16. A computer-readable medium as recited in claim 14, wherein instructions for receiving one or more phoneme substitution patterns further comprise instructions for receiving at least one phoneme substitution pattern that comprises (a) information identifying a first phoneme and (b) information indicating that the first phoneme is to be deleted from the dynamically linked phoneme values.

17. A computer-readable medium as recited in claim 14, wherein instructions for determining one or more best paths further comprise instructions for receiving two or more example pronunciations for a particular word under consideration and determining one or more best paths among the dynamically linked phoneme values for each of the example pronunciations.

18. A computer-readable medium as recited in claim 14, wherein instructions for determining one or more best paths further comprise instructions for receiving two or more example pronunciations for a particular word under consideration based on measurement data that is generated and stored by the speech recognition system during its execution.

19. A computer-readable medium as recited in claim 14, wherein instructions for determining one or more best paths further comprise instructions for converting each of the best paths into pronunciations according to a format of representing pronunciations that is used in the pronunciation dictionary.

20. A speech recognition apparatus comprising:

a storage medium having a pronunciation dictionary stored thereon, wherein each pronunciation in the pronunciation dictionary is represented as one or more dynamically linked phoneme values; and

an expansion mechanism and a best path mechanism communicatively coupled to the storage medium, wherein the expansion mechanism and the best path mechanism are configured to modify a pronunciation dictionary in a speech recognition system to include one or more alternate pronunciations by;

receiving one or more phoneme substitution patterns that define one or more substitute phoneme values for one or more of the dynamically linked phoneme values;

adding the substitute phoneme values and new links therefor to the dynamically linked phoneme values;

determining one or more best paths among the dynamically linked phoneme values for a particular word under consideration; and

modifying the pronunciation dictionary by adding a pronunciation that represents each of the best paths.