Unsupervised Unigram Discovery for Domain-Specific Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech recognition systems, such as LVCSR, are unable to recognize out-of-vocabulary words and domain-specific terms without extensive human intervention, limiting their effectiveness in specific domains like pharmaceutics.
Innovation Solution
An unsupervised method and system for automatically discovering unigrams by applying an acoustic machine-learning model, greedy decoder, and beam decoder to enhance language models, filtering and refining them with various filters to include new words and improve transcription accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a supervised learning mechanism is used to augment vocabulary by manually adding phrases to the language model, then the vocabulary can be updated and validated, but extensive human intervention is required
Solution Approach 1:
The system performs self-service by automatically discovering and validating new unigrams through unsupervised learning. The acoustic model processes speech data, the greedy decoder generates candidate unigrams, and the beam decoder validates them automatically without human intervention, enabling the system to expand its vocabulary autonomously
Solution Approach 2:
The system implements feedback mechanisms where the beam decoder evaluates candidate unigrams generated by the greedy decoder and feeds this information back to update the language model. This continuous feedback loop enables automatic validation and integration of new vocabulary items based on acoustic evidence
2Reliability
If a fixed vocabulary dictionary is used in LVCSR systems, then the system can recognize common words, but it cannot recognize out-of-vocabulary words or domain-specific terms
Solution Approach 1:
The system transitions from a static, fixed vocabulary to a dynamic, evolving vocabulary. The language model continuously updates by incorporating newly discovered unigrams through the automated discovery process, allowing the system to adapt to domain-specific terminology and out-of-vocabulary words in real-time
Solution Approach 2:
The system maintains continuous operation by continuously processing speech data, continuously discovering new unigrams, and continuously updating the language model. This uninterrupted cycle ensures the vocabulary always reflects the most recent and relevant domain-specific terms
3Adaptability or versatility
If domain-specific terms are excluded from global language models, then the model remains general-purpose, but the terms cannot be recognized in their specific field
Solution Approach 1:
The system achieves universality by maintaining a single language model that serves both general-purpose and domain-specific functions. The automated discovery process identifies domain-specific terms within general speech data and integrates them into the universal model, enabling the same model to handle both generic and specialized vocabulary without requiring separate domain-specific models
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A system and method of automatically discovering unigrams in a speech data element may include receiving a language model that includes a plurality of n-grams, where each n-gram includes one or more unigrams; applying an acoustic machine-learning (ML) model on one or more speech data elements to obtain a character distribution function; applying a greedy decoder on the character distribution function, to predict an initial corpus of unigrams; filtering out one or more unigrams of the initial corpus to obtain a corpus of candidate unigrams, where the candidate unigrams are not included in the language model; analyzing the one or more first speech data elements, to extract at least one n-gram that comprises a candidate unigram; and updating the language model to include the extracted at least one n-gram.