Dialog Device Utterance Length Prediction for End Point Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing dialog systems fail to accurately determine the end point of a user utterance, leading to increased waiting time and impairing the naturalness of conversation, as they do not consider the length attribute of the next utterance when selecting between acoustic and lexical feature models for end point determination.
Innovation Solution
A dialog device that predicts the length attribute of a user utterance and selects either an acoustic feature model or a lexical feature model to estimate the end point, using a prediction unit, selection unit, and estimation unit to reduce waiting time by determining the appropriate model for end point determination based on the predicted length attribute.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If an acoustic feature model is used for end point determination, then the processing speed is fast, but the accuracy for long utterances deteriorates
Solution Approach 1:
The system dynamically selects between acoustic feature model and lexical feature model based on the predicted length attribute of the user utterance. For short utterances, the acoustic feature model is used for fast processing. For long utterances, the lexical feature model is used for accurate end point determination. This dynamic adaptation resolves the contradiction between speed and accuracy.
2Measurement precision
If a lexical feature model is used for end point determination, then the accuracy is high, but the processing time increases
Solution Approach 1:
The system applies different feature models to different cases based on utterance length. The lexical feature model is applied locally to long utterances where accuracy is critical, while the acoustic feature model is applied to short utterances where speed is more important. This localized application resolves the contradiction between accuracy and processing time.
3Measurement precision
If the end point determination is delayed, then the model selection can be more accurate, but the user waiting time increases
Solution Approach 1:
The system performs preliminary action by predicting the length attribute of the user utterance before the utterance is fully completed. Based on this prediction, the appropriate feature model is selected in advance, allowing the system to process the utterance with the optimal model without delaying the end point determination. This resolves the contradiction between accuracy and user waiting time.
Data Source
AI summary
The dialog device according to the present invention includes a prediction unit 254 configured to predict an utterance length attribute of a user utterance in response to a the machine utterance, a selection unit 256 configured to use the utterance length attribute to select, as a feature model for usage in an end determination of the user utterance, at least one of an acoustic feature model or a lexical feature model, and an estimation unit 258 configured to estimate an end point in the user utterance using the selected model. By using this dialog device, it is possible to shorten the waiting time until a response is output to a user utterance by a machine, and to realize a more natural conversation between a user and a machine.