Dynamic Speech Model Selection for ASR Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic speech recognition systems perform poorly in varying environments and usage conditions due to the dominance of specific conditions over adaptation, leading to inappropriate models for multiple scenarios, especially when devices are shared among users and exposed to different background noises.
Innovation Solution
The method involves selecting a speaker-independent model and multiple speaker-dependent models based on available resources, recognizing utterances in parallel, and iteratively selecting a dominant model based on user profiles and environmental conditions, with clustering of utterances to generate condition-dependent models, and using beam search to determine the most effective model for recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a personalized speech model is provided for each speaker, then speech recognition accuracy is improved, but device complexity and resource consumption increase
Solution Approach 1:
The system dynamically selects between speaker-independent and speaker-dependent models based on current conditions rather than using a fixed approach. The model selection adapts to varying environments, user profiles, and computing resources available, resolving the contradiction by making the system flexible rather than static
Solution Approach 2:
The system changes the parameter of model selection based on multiple factors including user profile, environmental conditions, and computing resources. By adjusting which model to use based on these parameters, the system achieves high accuracy when needed while managing complexity through conditional selection
2Adaptability or versatility
If multiple speaker-dependent models are maintained for different conditions, then recognition performance in varying environments is improved, but memory and processing resources increase
Solution Approach 1:
The speaker-independent model serves as a universal baseline that can handle multiple conditions without requiring separate specialized models. This multi-functional approach allows the system to maintain environmental adaptability while reducing the total number of models needed
Solution Approach 2:
The system segments model usage by condition - using speaker-independent models for general conditions and speaker-dependent models only when specific conditions are met. This segmentation reduces overall storage requirements while maintaining versatility when needed
3Ease of manufacture
If conventional adaptation algorithms are used that assume consistent conditions, then model training is simplified, but recognition accuracy deteriorates in varying environments
Solution Approach 1:
The system dynamically adapts model selection based on environmental conditions rather than assuming consistency. This dynamic approach maintains training simplicity by using pre-trained models while achieving high accuracy in varying environments through conditional model selection
Solution Approach 2:
The system introduces an intermediary model selection layer that bridges the gap between simple conventional algorithms and complex adaptive systems. This intermediary selects appropriate models based on environmental conditions, maintaining training simplicity while improving accuracy
4Measurement precision
If speaker-dependent models are selected based on user profile, then personalization accuracy is improved, but computation time increases
Solution Approach 1:
The system performs preliminary actions by pre-configuring user profiles and associated models before runtime. During actual speech recognition, the system quickly matches the current user to their pre-configured profile and selects the appropriate model, achieving high personalization accuracy with minimal computation time
Solution Approach 2:
The system uses partial personalization by selecting from a limited set of pre-configured user profiles rather than performing full adaptive training at runtime. This partial approach achieves sufficient personalization accuracy while significantly reducing computation time
Data Source
AI summary
Disclosed herein are methods, systems, and computer-readable storage media for automatic speech recognition. The method includes selecting a speaker independent model, and selecting a quantity of speaker dependent models, the quantity of speaker dependent models being based on available computing resources, the selected models including the speaker independent model and the quantity of speaker dependent models. The method also includes recognizing an utterance using each of the selected models in parallel, and selecting a dominant speech model from the selected models based on recognition accuracy using the group of selected models. The system includes a processor and modules configured to control the processor to perform the method. The computer-readable storage medium includes instructions for causing a computing device to perform the steps of the method.