Speech analysis method

• US 20020026307A1
• Filed: 07/03/2001
• Published: 02/28/2002
• Est. Priority Date: 07/03/2000
• Status: Abandoned Application

• Current Assignee: Siemens AG
• Original Assignee: Siemens AG
• Inventors: Ruland, Tobias
• Application Number:
  US09/897,421
• Publication Number:
  US 20020026307A1
• Time in Patent Office: Days
• US Class Current: 704/9
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules

Abstract

A method for computer-supported speech analysis, in which a syntactic structure is assigned to an utterance. In this process, assignments are made with probabilities which depend on an expanded context.

Claims

1. A method for computer-supported speech analysis, in which a syntactic structure is assigned to an utterance, having

a context in the narrower sense for combinations of states and speech units, which is composed of speech categories, states, including resultant states, and actions,

an expanded context for the combinations of states and speech units which contains syntactic variables which are not contained in the context in the narrower sense,

and having the following steps

the utterance divided into the speech units,

the speech units are assigned to the speech categories,

a state is determined,

the state is combined with the speech category of a speech unit,

one or more actions are assigned to the combination of state and speech category with a probability which depends on the expanded context,

a number of resultant states is determined by carrying out the actions, and

the method is carried out again starting from the combination of the state with the speech category of a speech unit for at least one of the resultant states so that further speech units of the utterance are processed.

2. The method as claimed in claim 1, wherein the expanded context contains the dialogue act of the utterance.

3. The method as claimed in claim 1, wherein the expanded context contains the speech unit itself and/or further speech units of the utterance.

4. The method as claimed in claim 1, wherein the expanded context contains the speech style in which the speech unit and/or the utterance was spoken.

5. The method as claimed in claim 1, wherein an order is allocated to the speech units, and in that the speech units are processed in this allocated order.

6. The method as claimed in claim 5, wherein the allocated order corresponds to the order, or the inverted order of the speech units in the utterance.

7. The method as claimed in claim 1, wherein the expanded context is divided with respect to the syntactic variables into a plurality of subcontexts.

8. The method as claimed in claim 1, wherein the method is a stochastic parsing, in particular a stochastic LR parsing.

9. The method as claimed in claim 8, wherein one or more actions are assigned to a combination of state and speech category by a parsing table.

10. The method as claimed in claim 8, wherein the method has a stack.

11. The method as claimed in claim 10, wherein the expanded context contains an extreme speech category of the stack.

12. The method as claimed in claim 10, wherein the expanded context contains an extreme non-terminal speech category of the stack.

13. An system for computer-supported speech analysis, in which a syntactic structure is assigned to an utterance, having

a context in the narrower sense for combinations of states and speech units, which is composed of speech categories, states, including resultant states, and actions,

an expanded context for the combinations of states and speech units which contains syntactic variables which are not contained in the context in the narrower sense,

the system comprising:

a dividing unit to divide the utterance into the speech units,

a first assigning unit to assign the speech units to the speech categories,

a first determining unit to determine a state,

a combining unit to combine the state with the speech category of a speech unit,

a second assigning unit to assign one or more actions to the combination of state and speech category with a probability which depends on the expanded context,

a second determining unit to determine a number of resultant states is by carrying out the actions, and

a repeating unit to reactivate the combining unit, the second assigning unit and the second determining unit, for at least one of the resultant states so that further speech units of the utterance are processed.

14. (NEW) At least one computer readable medium storing at least one program for controlling at least one computer to perform a method in which a syntactic structure is assigned to an utterance, having

a context in the narrower sense for combinations of states and speech units, which is composed of speech categories, states, including resultant states, and actions,

an expanded context for the combinations of states and speech units which contains syntactic variables which are not contained in the context in the narrower sense,

the method comprising:

dividing the utterance into the speech units, assigning the speech units to the speech categories,

determining a state,

combining the state with the speech category of a speech unit,

assigning one or more actions to the combination of state and speech category with a probability which depends on the expanded context,

determining a number of resultant states is by carrying out the actions, and

repeating the method starting from the combination of the state with the speech category of a speech unit for at least one of the resultant states so that further speech units of the utterance are processed.