Voice Command Recognition via Acoustic Model Pronunciation Dictionary
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voice recognition technologies are unsuitable for devices with limited memory and no access to remote servers, as they require extensive user-specific training or remote processing, making them impractical for small footprint devices and IoT applications.
Innovation Solution
The method involves generating an acoustic model-specific statistical pronunciation dictionary using a large general speech dataset, constructing an augmented grammar candidate set to capture decoding variations, and using greedy or cross-entropy-based algorithms to improve voice command recognition accuracy by adding alternative expressions to the original grammar set.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional voice recognition technology is used, then recognition accuracy can be improved through extensive training or remote processing, but device complexity and memory requirements increase making it unsuitable for small footprint devices
Solution Approach 1:
The patent extracts and pre-computes acoustic model decoding variations into a separate pronunciation dictionary that can be stored independently. This dictionary captures common decoding paths and variations without requiring the full acoustic model to be present, allowing the recognition system to operate with reduced complexity while maintaining accuracy.
Solution Approach 2:
The patent performs preliminary analysis by running the acoustic model through a large general speech dataset beforehand to identify and store common decoding variations. This pre-computation creates a pronunciation dictionary that anticipates likely recognition scenarios, eliminating the need for extensive real-time processing during actual voice command recognition.
2Measurement precision
If extensive user-specific training is performed, then recognition accuracy improves, but training time and computational resources increase
Solution Approach 1:
The patent creates a pronunciation dictionary that serves multiple purposes: it captures acoustic model decoding variations, stores alternative pronunciations, and provides recognition patterns for various commands. This single pre-computed resource replaces the need for extensive user-specific training while maintaining high recognition accuracy across different users and scenarios.
3Measurement precision
If remote server processing is used, then recognition accuracy improves, but system independence and offline capability are reduced
Solution Approach 1:
The patent creates a simplified copy of the acoustic model's knowledge in the form of a pronunciation dictionary that can be stored locally on the device. This copy contains the essential recognition patterns and decoding variations needed for accurate voice command recognition, enabling the system to operate independently without remote server processing while maintaining recognition accuracy.
4Quantity of substance
If a small footprint acoustic model is used, then memory requirements are reduced, but recognition accuracy decreases due to inability to capture decoding variations
Solution Approach 1:
The patent introduces a pronunciation dictionary as an intermediary between the small footprint acoustic model and the voice command recognition process. This dictionary acts as a bridge that compensates for the limited capacity of the small model by providing pre-computed decoding variations and alternative pronunciations, thereby maintaining recognition accuracy without increasing memory usage significantly.
Data Source
AI summary
Various embodiments include methods and devices for implementing automatic grammar augmentation for improving voice command recognition accuracy in systems with a small footprint acoustic model. Alternative expressions that may capture acoustic model decoding variations may be added to a grammar set. An acoustic model-specific statistical pronunciation dictionary may be derived by running the acoustic model through a large general speech dataset and constructing a command-specific candidate set containing potential grammar expressions. Greedy based and cross-entropy-method (CEM) based algorithms may be utilized to search the candidate set for augmentations with improved recognition accuracy.