Related Experiment Videos
A multi-views multi-learners approach towards dysarthric speech recognition using multi-nets artificial neural
Summary
This study introduces a novel Dysarthric Multi-Networks Speech Recognizer (DM-NSR) to improve automatic speech recognition for individuals with dysarthria. The DM-NSR model significantly enhances speech recognition rates and reduces errors for dysarthric speakers.
Area of Science:
- Speech Processing
- Artificial Intelligence
- Neurology
Background:
- Dysarthria, a motor speech disorder, significantly impacts communication.
- Existing automatic speech recognition (ASR) systems perform poorly with dysarthric speech.
- Advanced ASR models are needed to accommodate speech variability in dysarthria.
Purpose of the Study:
- To develop an improved ASR model for individuals with dysarthria.
- To address the limitations of current ASR technologies in recognizing dysarthric speech.
- To enhance speech intelligibility and communication for dysarthric speakers.
Main Methods:
- Proposed a Dysarthric Multi-Networks Speech Recognizer (DM-NSR) utilizing a multi-nets artificial neural network approach.
- Implemented a multi-views multi-learners strategy to handle dysarthric speech variability.
- Evaluated both speaker-dependent and speaker-independent DM-NSR paradigms, comparing them against single-learner models and existing methods.
Main Results:
- The DM-NSR model demonstrated superior performance compared to legacy and reference models.
- Achieved an improved recognition rate by up to 24.67%.
- Reduced the error rate by up to 8.63%.
Conclusions:
- The proposed DM-NSR model effectively tolerates the complexities and variability of dysarthric speech.
- This approach offers a significant advancement in ASR technology for individuals with dysarthria.
- The DM-NSR shows promise for improving communication accessibility for dysarthric populations.