Aviation Speech Recognition With Contextual Vocabulary Biasing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic speech recognition (ASR) models struggle with accurate transcription of aviation speech due to the unique vocabulary and infrequent usage of aviation-related terms, which are not adequately represented in general language datasets.
Innovation Solution
Enhance ASR models with contextual biasing by leveraging aircraft state information and intent information to boost recognition of seldom-used aviation words and phrases, using data scraping from sources like ADS-B, ATIS, and flight plans to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If general language datasets are used for training ASR models, then the models can recognize common speech patterns, but they fail to accurately recognize aviation-specific vocabulary and terms
Solution Approach 1:
The patent applies local quality by creating context-specific vocabulary lists tailored to different aviation scenarios (e.g., approach, departure, cruise). Instead of using a uniform general language model, the system adapts the vocabulary weights locally based on the current flight context, aircraft type, and operational phase, thereby improving recognition accuracy for aviation-specific terms while maintaining general speech recognition capabilities
Solution Approach 2:
The system dynamically changes the parameter of vocabulary weighting based on contextual information. By adjusting the weights of specific words and phrases according to the current aviation context (such as flight phase, aircraft type, or operational mode), the ASR model can adapt its recognition sensitivity to match the expected terminology, resolving the contradiction between general speech patterns and aviation-specific vocabulary
2Measurement precision
If aviation-specific vocabulary is added to the ASR model, then recognition of aviation terms improves, but the complexity of the system increases
Solution Approach 1:
The patent applies preliminary action by pre-generating context-specific vocabulary lists before speech recognition occurs. These vocabulary lists are created based on expected flight contexts, aircraft types, and operational phases, allowing the ASR model to have aviation-specific terms ready in advance rather than processing them in real-time, thus improving accuracy without significantly increasing system complexity during operation
Solution Approach 2:
The system introduces an intermediary component that acts as a bridge between the general ASR model and aviation-specific terminology. This intermediary layer manages the vocabulary weights and context information, allowing the core ASR model to remain relatively simple while still achieving accurate aviation term recognition through the mediating vocabulary adjustment mechanism
3Measurement precision
If context-specific vocabulary weighting is implemented, then speech recognition accuracy for aviation speech improves, but the processing time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing context-specific vocabulary lists that are generated based on expected flight scenarios. These pre-computed vocabulary weights are readily available during speech recognition, eliminating the need for real-time computation of contextual weights and thereby reducing processing time while maintaining improved accuracy
Solution Approach 2:
The system applies partial action by selectively weighting only the most relevant vocabulary terms for the current context rather than adjusting all vocabulary weights uniformly. This selective approach focuses computational resources on the most critical aviation-specific terms, improving recognition accuracy for key terminology while minimizing the overall computational burden and processing time
Data Source
AI summary
A variety of applications can include a system having a speech recognition system responsive to the speech input, where the speech recognition system can be configured to recognize the speech input using an aviation vocabulary including words extracted using state information of an aircraft or intent information of the aircraft associated with the received speech input. A control system can be implemented to automatically perform an action in the system in response to analysis of the recognized speech input, where the action is associated with flight of the aircraft.