Multi-Engine Speech Transcription via Dynamic Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic speech recognition systems rely on a single speech recognition model, which limits their ability to accurately transcribe multiple languages and speech qualities, leading to trade-offs in accuracy and efficiency.
Innovation Solution
The system generates speech recognition model specifications for multiple distinct speech-to-text engines, each corresponding to a specific speech recognition model. It segments audio recordings into phrases, determines the appropriate engine based on pre- and post-transcription evaluation parameters, and accepts hypotheses to improve transcription accuracy and speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single speech recognition model is used, then the system complexity is low, but the transcription accuracy across multiple languages and speech qualities deteriorates
Solution Approach 1:
The patent segments the audio recording into multiple audio segments, each corresponding to a single phrase of a single speaker. This segmentation allows the system to apply different speech recognition models to different segments based on their specific characteristics (language, accent, audio quality), thereby improving overall transcription accuracy while managing system complexity through targeted model selection rather than universal application of all models
Solution Approach 2:
The system dynamically selects which speech recognition model to apply to each audio segment based on pre-transcription evaluation parameters (language, accent, audio quality) and post-transcription evaluation parameters (confidence scores, hypothesis quality). This dynamic adaptation allows the system to optimize transcription accuracy for each segment while avoiding the constant complexity of running all models simultaneously
2Measurement precision
If multiple speech recognition models are evaluated for each audio segment, then the transcription accuracy improves, but the processing time increases
Solution Approach 1:
The system performs preliminary evaluation of audio segments using pre-transcription evaluation parameters (language, accent, audio quality) to determine which speech recognition models are most likely to succeed. This preliminary action filters out unsuitable models before full transcription, reducing the number of models that need to be fully evaluated while maintaining high transcription accuracy
Solution Approach 2:
The system uses post-transcription evaluation parameters (confidence scores, hypothesis quality) as feedback to determine whether to accept the current transcription or submit the audio segment to additional models. This feedback mechanism prevents unnecessary processing time by stopping the evaluation process early when satisfactory results are achieved, while still allowing for improved accuracy when the feedback indicates room for enhancement
Data Source
AI summary
In some embodiments, an exemplary inventive system for improving computer speed and accuracy of automatic speech transcription includes at least components of: a computer processor configured to perform: generating a recognition model specification for a plurality of distinct speech-to-text transcription engines; where each distinct speech-to-text transcription engine corresponds to a respective distinct speech recognition model; receiving at least one audio recording representing a speech of a person; segmenting the audio recording into a plurality of audio segments; determining a respective distinct speech-to-text transcription engine to transcribe a respective audio segment; receiving, from the respective transcription engine, a hypothesis for the respective audio segment; accepting the hypothesis to remove a need to submit the respective audio segment to another distinct speech-to-text transcription engine, resulting in the improved computer speed and the accuracy of automatic speech transcription and generating a transcript of the audio recording from respective accepted hypotheses for the plurality of audio segments.