Speech Recognition Field Identification via Frequency-Based Language Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition technologies face challenges in accurately identifying the field of input speech when it is unknown, leading to increased processing loads due to the need for multiple language models and manual classification of speech files.
Innovation Solution
A speech recognition apparatus and method that generates an identification language model by selecting phrases with high appearance frequency on a field-by-field basis, allowing for automatic field determination and reduced processing load by using a single identification language model for initial recognition, followed by customization of the language model for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple language models are prepared to handle unknown field speech, then the ability to recognize speech from unknown fields is improved, but the processing load increases excessively
Solution Approach 1:
The patent segments the language model selection process into two stages: first using a general-purpose language model for initial recognition, then selectively switching to field-specific language models only when needed. This segmentation allows the system to handle unknown field speech without continuously operating multiple specialized models, thus reducing processing load while maintaining adaptability.
Solution Approach 2:
The patent implements dynamic language model selection based on field identification results. The system transitions from a static configuration of multiple fixed language models to a dynamic approach where the appropriate language model is selected and switched based on real-time field detection, optimizing processing resources according to actual speech content.
2Measurement precision
If multiple language models are used to improve recognition accuracy for specific fields, then recognition accuracy for known fields is improved, but the complexity of the system increases
Solution Approach 1:
The patent employs a universal general-purpose language model that can handle all fields as its primary function, while field-specific language models serve as supplementary specialized functions. This multi-functionality approach allows the system to maintain high recognition accuracy for known fields through specialization while avoiding the complexity of requiring separate dedicated models for every possible field.
Solution Approach 2:
The patent performs preliminary field identification using a general-purpose language model before engaging field-specific language models. This preliminary action filters out the need for multiple specialized models in most cases, reducing system complexity while preserving the ability to achieve high accuracy when field-specific recognition is required.
3Ease of manufacture
If manual classification of speech files by field is performed in advance, then the organization of speech data is improved, but the time and labor required increase
Solution Approach 1:
The patent implements automatic field identification where the speech recognition system itself performs the classification function by detecting the field from the speech content in real-time. This self-service approach eliminates the need for manual pre-classification of speech files, significantly reducing time and labor requirements while maintaining effective organization of speech data by field.
Data Source
AI summary
A speech recognition apparatus 20 includes: an identification language model creation unit 21 that selects, from learning texts 27 for various fields for generating language models 26 for the fields, a phrase that includes a word whose appearance frequency satisfies a set condition on a field-by-field basis, and generates an identification language model 25 for identifying the field of speech using the selected phrases; a speech recognition unit 22 that executes speech recognition on the speech using the identification language model 25, and outputs text data and word confidences as a recognition result; and a field determination unit 23 that specifies a field that includes the most words whose confidences are greater than or equal to a set value based on the text data, the word confidences, and the words in the learning texts for the fields, and determines that the specified field is the field of the speech.