CONFIDENCE CALIBRATION IN AUTOMATIC SPEECH RECOGNITION SYSTEMS
First Claim
1. A system comprising, a calibration model, the calibration model configured to receive a confidence score and associated word from a speech recognition engine, and adjust the confidence score to provide a calibrated confidence score for use by an application, the calibration model having been trained for a usage scenario based upon a calibration training set obtained from at least one previous corresponding usage scenario.
2 Assignments
0 Petitions
Accused Products
Abstract
Described is a calibration model for use in a speech recognition system. The calibration model adjusts the confidence scores output by a speech recognition engine to thereby provide an improved calibrated confidence score for use by an application. The calibration model is one that has been trained for a specific usage scenario, e.g., for that application, based upon a calibration training set obtained from a previous similar/corresponding usage scenario or scenarios. Different calibration models may be used with different usage scenarios, e.g., during different conditions. The calibration model may comprise a maximum entropy classifier with distribution constraints, trained with continuous raw confidence scores and multi-valued word tokens, and/or other distributions and extracted features.
50 Citations
20 Claims
- 1. A system comprising, a calibration model, the calibration model configured to receive a confidence score and associated word from a speech recognition engine, and adjust the confidence score to provide a calibrated confidence score for use by an application, the calibration model having been trained for a usage scenario based upon a calibration training set obtained from at least one previous corresponding usage scenario.
- 14. In a computing environment, a method performed on at least one processor, comprising, training a calibration model for use in adjusting confidence scores output by a speech recognizer, including processing a calibration training set containing words, confidence scores, and labels indicating whether each word was correctly recognized, extracting features from the calibration training set including word and score distribution features, and using the features to train the calibration model.
-
20. One or more computer-readable media having computer-executable instructions, which when executed perform steps, comprising:
-
training a calibration model, including processing a calibration training set containing data obtained from one or more previous usage scenarios; receiving a confidence score from a speech recognition engine at the calibration model, and adjusting the confidence score to output a calibrated confidence score for a usage scenario that corresponds to the one or more previous usage scenarios.
-
Specification