Speech Recognition Key Phrase Decoupling via Independent Probability Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition technologies face challenges in accurately recognizing key phrases due to the influence of traditional language models, which often misinterpret key phrases as other phrases or brand names, leading to inaccurate determination of query commands.
Innovation Solution
The method involves adding a key phrase to a dictionary with independent probabilities for each dictionary phrase, using a N-gram language model that decouples the key phrase from the probabilities of subsequent phrases, allowing for more accurate recognition by relying on preceding phrases and sentence start probabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a traditional language model is used for speech recognition, then the system can process general language sequences, but the key phrase accuracy deteriorates due to misinterpretation as other phrases or brand names
Solution Approach 1:
The patent segments the speech recognition process into two distinct parts: key phrase recognition and subsequent phrase recognition. The key phrase is processed separately with dedicated probability calculations, while following phrases use standard language model probabilities. This segmentation prevents the key phrase from being misinterpreted by the general language model.
Solution Approach 2:
The patent applies different probability calculation methods to different parts of the speech sequence. The key phrase position receives special treatment with independent probability scoring, while subsequent phrases use standard N-gram probabilities. This local differentiation ensures high accuracy for the critical key phrase while maintaining overall language fluency.
2Stability of the object's composition
If the key phrase is included in standard probability calculations with preceding phrases, then the language model can maintain grammatical coherence, but the key phrase probability score deteriorates due to dependence on uncommon preceding phrases
Solution Approach 1:
The patent segments the probability calculation into two independent parts: the key phrase probability is calculated separately without dependence on preceding phrases, while the subsequent phrases maintain their standard N-gram probability calculations based on linguistic context. This ensures the key phrase score reflects its actual occurrence frequency rather than being penalized by uncommon preceding phrases.
Solution Approach 2:
The patent extracts the key phrase probability calculation from the standard N-gram language model sequence. By taking out the key phrase probability computation and calculating it independently, the system eliminates the harmful dependence on preceding phrases while maintaining the overall language model structure for subsequent phrases.
3Device complexity
If the system uses standard N-gram probability for all phrases including the key phrase, then the processing is simple and unified, but the determination of query commands deteriorates due to errors in probability scoring
Solution Approach 1:
The patent segments the speech sequence into a key phrase portion and a subsequent phrases portion, applying different probability calculation strategies to each. This segmentation improves query command identification accuracy by ensuring the key phrase probability is not distorted by standard N-gram dependencies, while maintaining reasonable system complexity through the use of established N-gram methods for the majority of the speech sequence.
Data Source
AI summary
Systems and methods are provided for speech recognition. An example method may be implementable by a server. The method may comprise adding a key phrase into a dictionary comprising a plurality of dictionary phrases, and for each one or more of the dictionary phrases, obtaining a first probability that the dictionary phrase is after the key phrase in a phrase sequence. The key phrase and the dictionary phrase may each comprise one or more words. The first probability may be independent of the key phrase.