ASR Model Training With Session-by-Session Data Deletion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current audio transcription systems for hard-of-hearing or deaf individuals suffer from inaccuracies and delays due to reliance on human revoicing and speaker-dependent automatic speech recognition (ASR) systems, leading to increased costs and reduced effectiveness.
Innovation Solution
Implement a hybrid system combining speaker-dependent and speaker-independent ASR systems with real-time transcription fusion and human revoicing, utilizing a transcription unit that includes multiple ASR systems and a fuser to generate accurate and synchronized transcriptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If human revoicing is used for transcription, then transcription accuracy may be improved, but costs increase and delays occur
Solution Approach 1:
The system performs preliminary transcription using ASR models before human review, preparing multiple candidate transcriptions in advance. This allows human reviewers to focus only on correcting errors rather than creating transcriptions from scratch, reducing both time and cost while maintaining accuracy.
Solution Approach 2:
The system introduces an intermediary layer that combines ASR model outputs with human feedback. The ASR models generate initial transcriptions, and human reviewers provide corrective feedback, which is then used to refine the models. This intermediary process resolves the contradiction by automating routine tasks while preserving human expertise for critical corrections.
2Measurement precision
If speaker-dependent ASR systems are used, then transcription accuracy improves, but device complexity and costs increase
Solution Approach 1:
The system employs speaker-independent ASR models that can handle multiple speakers without requiring separate models for each user. These universal models reduce device complexity and training requirements while maintaining acceptable accuracy through continuous learning from aggregated feedback across multiple users and sessions.
Solution Approach 2:
The system dynamically adjusts model parameters based on feedback from human reviewers and transcription accuracy metrics. By changing parameters such as language model weights and acoustic model configurations based on performance data, the system maintains high accuracy without requiring complex speaker-specific adaptations for each user.
3Measurement precision
If multiple ASR systems are combined in a hybrid system, then transcription accuracy improves, but device complexity increases
Solution Approach 1:
The system merges multiple ASR models and human feedback mechanisms into a unified transcription pipeline. By combining speaker-independent models with continuous learning capabilities and human review processes, the system achieves high accuracy while managing complexity through integrated architecture that shares common components across different models.
Data Source
AI summary
A method may include obtaining first audio data of a first communication session between a first and second device and during the first communication session, obtaining a first text string that is a transcription of the first audio data and training a model of an automatic speech recognition system using the first text string and the first audio data. The method may further include in response to completion of the training, deleting the first audio data and the first text string and after deleting the first audio data and the first text string, obtaining second audio data of a second communication session between a third and fourth device and during the second communication session obtaining a second text string that is a transcription of the second audio data and further training the model of the automatic speech recognition system using the second text string and the second audio data.