Unsupervised training in natural language call routing
First Claim
1. A method of training a natural language call routing system using an unsupervised trainer, the call routing system being adapted to associate a unique identifier to an incoming call and having a first speech recognizer adapted to recognize speech input from a caller and to output text corresponding to the input speech, a topic classifier adapted to identify a topic from the text output by the first speech recognizer, and a call router adapted to route the caller to an agent or an automated fulfillment in accordance with the identified topic, the unsupervised trainer being adapted to tune performance of the call routing system, the method of training comprising:
- storing, in a waveform database, audio data from an incoming call and associated unique identifier information for the incoming call;
applying a second speech recognizer to the audio data from the waveform database to produce a text transcription of the stored audio for the call;
forwarding outputs of the second speech recognizer to a training database, the training database being adapted to store text transcripts from the second recognizer with respective unique call identifiers as well as topic data;
for a call routed by the call router to an agent;
entering a call topic determined by the agent into a form; and
supplying the call topic information from the form to the training database together with the associated unique call identifier; and
for a call routed to automated fulfillment;
querying the caller regarding the true topic of the call; and
adding this topic information, together with the associated unique call identifier, to the training database; and
performing topic identification model training and statistical grammar model training on the basis of the topic information and transcription information stored in the training database.
8 Assignments
0 Petitions
Accused Products
Abstract
A method of training a natural language call routing system using an unsupervised trainer is provided. The unsupervised trainer is adapted to tune performance of the call routing system on the basis of feedback and new topic information. The method of training comprises: storing audio data from an incoming call as well as associated unique identifier information for the incoming call; applying a highly accurate speech recognizer to the audio data from the waveform database to produce a text transcription of the stored audio for the call; forwarding outputs of the second speech recognizer to a training database, the training database being adapted to store text transcripts from the second recognizer with respective unique call identifiers as well as topic data; for a call routed by the call router to an agent: entering a call topic determined by the agent into a form; and supplying the call topic information from the form to the training database together with the associated unique call identifier; and for a call routed to automated fulfillment: querying the caller regarding the true topic of the call; and adding this topic information, together with the associated unique call identifier, to the training database; and performing topic identification model training and statistical grammar model training on the basis of the topic information and transcription information stored in the training database.
249 Citations
29 Claims
-
1. A method of training a natural language call routing system using an unsupervised trainer, the call routing system being adapted to associate a unique identifier to an incoming call and having a first speech recognizer adapted to recognize speech input from a caller and to output text corresponding to the input speech, a topic classifier adapted to identify a topic from the text output by the first speech recognizer, and a call router adapted to route the caller to an agent or an automated fulfillment in accordance with the identified topic, the unsupervised trainer being adapted to tune performance of the call routing system, the method of training comprising:
-
storing, in a waveform database, audio data from an incoming call and associated unique identifier information for the incoming call; applying a second speech recognizer to the audio data from the waveform database to produce a text transcription of the stored audio for the call; forwarding outputs of the second speech recognizer to a training database, the training database being adapted to store text transcripts from the second recognizer with respective unique call identifiers as well as topic data; for a call routed by the call router to an agent; entering a call topic determined by the agent into a form; and supplying the call topic information from the form to the training database together with the associated unique call identifier; and for a call routed to automated fulfillment; querying the caller regarding the true topic of the call; and adding this topic information, together with the associated unique call identifier, to the training database; and performing topic identification model training and statistical grammar model training on the basis of the topic information and transcription information stored in the training database. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
-
-
10. A system for unsupervised training of a natural language call routing system, the call routing system being adapted to associate a unique identifier to an incoming call and having a first speech recognizer adapted to recognize speech input from a caller and to output text corresponding to the input speech, a topic classifier adapted to identify a topic from the text output by the first speech recognizer, and a call router adapted to route the caller to an agent or an automated fulfillment in accordance with the identified topic, the system for unsupervised training being adapted to tune performance of the call routing system, the unsupervised training system comprising:
-
a waveform database, adapted to store audio data from an incoming call and associated unique identifier information for the incoming call; a second speech recognizer adapted to be applied to the audio data from the waveform database to produce a text transcription of the stored audio for the call; a training database, the training database adapted to store text transcripts from the second recognizer with respective unique call identifiers as well as topic data; an online topic annotator including a form adapted to allow entry of a call topic determined by the agent for a call routed to the agent and, and a true topic entry unit for entry of a call topic gathered from querying the caller regarding the true topic of a call, for a call routed to automated fulfillment, the annotated topic information, together with the associated unique call identifier, being supplied to the training database together with the associated unique call identifier; a topic trainer adapted to perform topic identification model training on the basis of the topic information and transcription information stored in the training database; and a statistical grammar builder adapted to perform grammar model training on the basis of the topic information and transcription information stored in the training database. - View Dependent Claims (11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A trainer for unsupervised training of a natural language call routing system, the call routing system being adapted to associate a unique identifier to an incoming call and having first speech recognizing means for recognizing speech input from a caller and outputting text corresponding to the input speech, topic classifying means for identifying a topic from the text output by the first speech recognizing means, and call routing means for routing the caller to an agent or an automated fulfillment in accordance with the identified topic, the trainer being adapted to tune performance of the call routing system, the trainer comprising:
-
a waveform database, adapted to store audio data from an incoming call as well as associated unique identifier information for the incoming call; second speech recognizing means for producing, from the audio data from the waveform database, a text transcription of the stored audio for the call; a training database, the training database adapted to store text transcripts from the second recognizing means with respective unique call identifiers as well as topic data; online topic annotation means for providing to an agent a form adapted to allow entry of a call topic determined by the agent for a call routed to the agent and, and including a true topic entry unit for entry of a call topic gathered from querying the caller regarding the true topic of a call, for a call routed to automated fulfillment, the annotated topic information, together with the associated unique call identifier, being supplied to the training database together with the associated unique call identifier; topic training means for performing topic identification model training on the basis of the topic information and transcription information stored in the training database; and statistical grammar building means for performing grammar model training on the basis of the topic information and transcription information stored in the training database. - View Dependent Claims (20, 21, 22, 23, 24, 25, 26, 27)
-
-
28. A method of training a speech recognition system using an unsupervised trainer, the speech recognition system being adapted to associate a unique identifier to an incoming call and having a first speech recognizer adapted to recognize speech input from a caller and to output text corresponding to the input speech, and an associated call router which routes the caller to an agent or an automated fulfillment, using some or all of the information provided by the caller, the unsupervised trainer being adapted to tune performance of the speech recognition system on the basis of feedback using both automatic and manual transcription, the method of training comprising:
-
storing, in a waveform database, audio data from the incoming call as well as associated unique identifier information for the incoming call; applying a second speech recognizer to the audio data from the waveform database to produce a text transcription of the stored audio data for the call; forwarding outputs of the second speech recognizer to a training database, the training database being adapted to store text transcripts from the second recognizer with respective unique call identifiers as well as topic data; agents entering information which is being captured by the speech recognition system, determined by the agent, into a form; supplying the information from the form to the training database together with the associated unique call identifier; and performing voice model training on the basis of the transcription and audio information stored in the training database.
-
-
29. A system for unsupervised training of a speech recognition system, the speech recognition system being adapted to associate a unique identifier to an incoming call and having a first speech recognizer adapted to recognize speech input from a caller and to output text corresponding to the input speech, and a call router adapted to route the caller to an agent or an automated fulfillment in accordance with information provided by the caller, the unsupervised training system being adapted to tune performance of the speech recognition system on the basis of feedback, the unsupervised training system comprising:
-
a waveform database, adapted to store audio data from the incoming call as well as associated unique identifier information for the incoming call; a second speech recognizer adapted to be applied to the audio data from the waveform database to produce a text transcription of the stored audio data for the call; a training database adapted to store text transcripts from the second recognizer with respective unique call identifiers; an online annotator including a form adapted to allow entry of true information determined by the agent, for a call routed to the agent, the annotated topic information, together with the associated unique call identifier, being supplied to the training database together with the associated unique call identifier; and a voice model trainer adapted to perform voice model training on the basis of the transcription and audio information stored in the training database.
-
Specification