Monitoring, mining, and classifying electronically recordable conversations

• US 20070067159A1
• Filed: 09/02/2005
• Published: 03/22/2007
• Est. Priority Date: 09/02/2005
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Basu, Sumit, de la Fuente, Mauricio
• Granted Patent: US 7,596,498 B2
• Time in Patent Office: Days
• US Class Current: 704/200
• CPC Class Codes: G10L 17/00 Speaker identification or verification techniques
• CPC Class Codes: G10L 25/27 characterised by the analysis technique
• CPC Class Codes: G10L 25/51 for comparison or discrimination
• CPC Class Codes: G10L 25/60 for measuring the quality of voice signals
• CPC Class Codes: H04M 2203/301 Management of recordings
• CPC Class Codes: H04M 2203/559 Sorting systems
• CPC Class Codes: H04M 3/2281 Call monitoring, e.g. for law enforcement purposes; Call tracing; Detection or prevention of malicious calls
• CPC Class Codes: H04M 3/42221 Conversation recording systems at the subscriber's set H04M1/656
• CPC Class Codes: H04M 3/51 Centralised call answering arrangements requiring operator intervention , e.g. call or contact centers for telemarketing

Abstract

Conversations that take place over an electronically recordable channel are analyzed by constructing a set of features from the speech of two participants in the conversation. The set of features is applied to a model or a plurality of models to determine the likelihood of the set of features for each model. These likelihoods are then used to classify the conversation into categories, provide real-time monitoring of the conversation, and/or identify anomalous conversations.

Claims

1. A method of analyzing a conversation over an electronically recordable channel, the method comprising:

constructing a feature from speech of two participants in the conversation;

applying the feature to a model to determine a likelihood for the feature; and

indicating whether the conversation matches a model based on the likelihood for the feature.

2. The method of claim 1 wherein in the model is trained based on a plurality of conversations, wherein each conversation has a same person as a participant.

3. The method of claim 1 wherein indicating whether the conversation matches a model comprises indicating whether the conversation matches a model to a monitoring application that is used to monitor a plurality of conversations at the same time.

4. The method of claim 3 wherein the monitoring application provides visual indications for a plurality of calls.

5. The method of claim 3 wherein indicating whether the conversation matches a model comprises indicating whether the conversation matches a model to one of the participants of the conversation.

6. The method of claim 1 wherein constructing a feature comprises constructing a short term feature representing only a portion of the conversation and wherein indicating whether the conversation matches a model comprises indicating whether a short term portion of the conversation matches the model.

7. The method of claim 1 wherein constructing a feature comprises constructing a long term feature that represents the entire conversation and wherein indicating whether the conversation matches a model comprises indicating whether the conversation as a whole matches the model.

8. The method of claim 1 wherein indicating whether the conversation matches a model comprises storing an indication of whether the conversation matches the model together with a representation of the contents of the conversation.

9. A computer-readable medium having computer-executable instructions for performing steps comprising:

generating a feature that is based on speech signals from at least two participants in a conversation held over an electronically recordable channel; and

using the feature to provide a description of the conversation.

10. The computer-readable medium of claim 9 wherein using the feature to provide a description of the conversation comprises displaying an item that is representative of the value of the feature.

11. The computer-readable medium of claim 9 wherein using the feature to provide a description of the conversation comprises using the feature to classify the conversation into one of a number of categories.

12. The computer-readable medium of claim 11 wherein using the feature to classify the conversation comprises applying a feature to at least one model to classify the conversation.

13. The computer-readable medium of claim 11 wherein applying the a feature to at least one model comprises applying the feature to a plurality of models, identifying a model that provides a highest likelihood for the feature, and selecting a class associated with the identified model as the class for the conversation.

14. The computer-readable medium of claim 12 wherein applying the feature to at least one model comprises applying the feature to a model of a desired conversation, determining how the feature deviates from the model of the desired conversation, and classifying the conversation based on how the feature deviates from the model of the desired conversation.

15. The computer-readable medium of claim 11 further comprising providing an indication of the classification of the conversation to a monitoring application.

16. The computer-readable medium of claim 15 wherein the monitoring application provides visual indications of the classifications of a plurality of conversations.

17. The computer-readable medium of claim 16 wherein the visual indications of the classifications of the plurality of conversations are provided while the conversations are in progress.

18. A method comprising:

classifying a conversation that occurred between two participants over an electronically recordable channel based on speech signals from both participants without reference to the words represented by the speech signals;

storing the classification and a representation of the content of the conversation in a storage area;

searching the storage area for conversations that have a particular classification; and

retrieving representations of the content of the conversations found during the search.

19. The method of claim 18 wherein classifying a conversation comprises deriving a feature from the speech signals and applying the feature to a model of a conversation.

20. The method of claim 18 wherein searching for conversations further comprises searching for conversations related to a particular entity.