Language Model Selection for Accurate Closed Captioning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems face challenges in accurately detecting languages and generating closed captioning due to variations in speech patterns and geographic pronunciation, which affect the accuracy of captioning text displayed with streaming videos.
Innovation Solution
A method that isolates an audio stream, identifies and selects a language model based on the audio stream, creates text from the selected model, generates a confidence level for the acoustic model, and updates the language model through online learning to improve captioning accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single language model is used for speech recognition, then the system complexity is reduced, but the accuracy of closed captioning decreases due to variations in speech patterns and geographic pronunciation
Solution Approach 1:
The patent segments the language model into multiple specialized models, each trained on specific geographic or dialectal speech patterns. The system divides the task of speech recognition across multiple language models rather than using a single general model, allowing each model to specialize in particular pronunciation variations and speech characteristics.
Solution Approach 2:
The patent applies local quality by assigning different language models to different geographic regions or speech patterns. Each language model is optimized for local speech characteristics, and the system selects the appropriate local model based on detected speech patterns, ensuring that each region receives specialized recognition capabilities rather than a one-size-fits-all approach.
2Measurement precision
If multiple language models are identified and selected based on audio stream analysis, then the captioning accuracy is improved, but the device complexity and processing time increase
Solution Approach 1:
The patent applies preliminary action by pre-training and organizing multiple language models before runtime, and by performing preliminary analysis of the audio stream to identify speech characteristics. The system prepares language models in advance and conducts initial speech pattern detection to facilitate faster model selection during actual speech recognition operations.
Solution Approach 2:
The patent implements feedback mechanisms where the system continuously monitors speech recognition results and adjusts language model selection based on performance metrics. The confidence level calculations provide feedback that informs whether to switch between language models, allowing the system to learn from its recognition accuracy and optimize model selection over time.
3Reliability
If language models are continuously refined through online learning, then the captioning proficiency improves over time, but the processing time and computational resources increase
Solution Approach 1:
The patent applies periodic action by implementing online learning at specific intervals or under certain conditions rather than continuously. The system periodically updates language models with new speech data and re-evaluates model performance, balancing the need for improved proficiency with the constraint of processing time by not constantly retraining models.
Data Source
AI summary
In an approach for acoustic modeling with a language model, a computer isolates an audio stream. The computer identifies one or more language models based at least in part on the isolated audio stream. The computer selects a language model from the identified one or more language models. The computer creates a text based on the selected language model and the isolated audio stream. The computer creates an acoustic model based on the created text. The computer generates a confidence level associated with the created acoustic model. The computer selects a highest ranked language model based at least in part on the generated confidence level.