Site News  |   Monitor Keywords  |   Monitor Archive  |   Organizer  |   Account Info  |

Hierarchical methods and apparatus for extracting user intent from spoken utterances

Hierarchical methods and apparatus for extracting user intent from spoken utterances description/claims

This application is a continuation of U.S. application Ser. No. 11/216,483 filed on Aug. 31, 2005, the disclosure of which is incorporated herein by reference.

The present invention relates generally to speech processing systems and, more particularly, to systems for hierarchically extracting user intent from spoken utterances, such as spoken instructions or commands.

The use of a speech recognition system (or a voice system) to translate a user's spoken command to a precise text command that the target system can input and process is well known. For example, in a conventional voice system based in a vehicle, a user (e.g., driver) interacts with the voice system by uttering very specific commands that must be consistent with machine-based grammar that is understood by the target system.

By way of example, assume that the climate control system in the vehicle is the target system. In order to decrease the temperature in the vehicle, the user of a conventional voice system may typically have to utter several predetermined machine-based grammar commands, such as the command “climate control” followed by the command “air conditioner” followed by the command “decrease temperature” followed by the command “five degrees.”

Unfortunately, people do not talk or think in terms of specific machine-based grammar, and may also forget the precise predetermined commands that must be uttered to effectuate their wishes.

One approach that attempts to overcome the machine-based grammar problem is to use a single-stage front end action classifier that detects a very general subject from the user's speech, which is then provided to a human operator for further intent determination. This is typically the approach used in the General Motors' OnStar™ system. However, a major problem with this approach is that a human operator is required.

Another approach is to build a full-fledged statistical parser, which takes the input as transcribed and builds a parse tree which is mined later to extract intent. One major difficulty in this second approach is that statistical parsers are huge in terms of storage requirements. Further, they require hand-tuning in every step. That is, every time data is added, the statistical parser requires a tremendous amount of hand-tuning and balancing of the new data with the old data.

Accordingly, improved techniques are needed that permit a user to employ more human-based grammar (i.e., free form or conversational input) while addressing a target system via a voice system.

Principles of the present invention provide improved techniques for permitting a user to employ more human-based grammar (i.e., free form or conversational input) while addressing a target system via a voice system.

In one aspect of the invention, a technique for determining intent associated with a spoken utterance of a user comprises the following steps/operations. Decoded speech uttered by the user is obtained. An intent is then extracted from the decoded speech uttered by the user. The intent is extracted in an iterative manner such that a first class is determined after a first iteration and a sub-class of the first class is determined after a second iteration. The first class and the sub-class of the first class are hierarchically indicative of the intent of the user, e.g., a target and data that may be associated with the target.

The multi-stage intent extraction approach may have more than two iterations. By way of example only, the user intent extracting step may further determine a sub-class of the sub-class of the first class after a third iteration, such that the first class, the sub-class of the first class, and the sub-class of the sub-class of the first class are hierarchically indicative of the intent of the user.

In a preferred embodiment, as will be explained in further detail below, the first class may represent a target (e.g., topic) associated with the user intent, the sub-class of the first class may represent an action (e.g., function) associated with the target, and the sub-class of the sub-class of the first class may represent data associated with the action. One or more commands may then be provided to a target system based on the class and sub-class determinations.

These and other objects, features and advantages of the present invention will become apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings.

• Provisional Patent
• Utility Patent

• What Is a Patent?
• What Is a Trademark or Servicemark?
• What Is a Copyright?
• Patent Laws