Related Experiment Video
Updated: Jun 26, 2026
05:48
Memorization-Based Training and Testing Paradigm for Robust Vocal Identity Recognition in Expressive Speech Using Event-Related Potentials Analysis
Published on: August 9, 2024
Adaptive Phoneme State Learning Architecture for Enhanced Speech Recognition Using Backpropagation Neural Network and
Rashmi Siddalingappa1, Deepa S2, Margaret Savitha2
1Computer and Data Science, York St John University, London, England, E14 2BA, UK.
F1000Research
|June 25, 2026
Summary
This study introduces an Adaptive Phoneme State Learning (APSL) algorithm to enhance automated speech recognition (ASR) systems. The APSL-BPNN-HMM model significantly improves accuracy and reduces Word Error Rate (WER) in speech-to-text conversion.
Area of Science:
- Computer Science
- Artificial Intelligence
- Signal Processing
Background:
- Automated speech recognition (ASR) systems struggle with accent variability, noise, and data privacy.
- Existing ASR architectures require enhancements for robust performance across diverse conditions.
Purpose of the Study:
- To propose and evaluate an enhanced ASR architecture using an Adaptive Phoneme State Learning (APSL) algorithm.
- To improve phoneme transition modeling and alignment for more accurate speech-to-text conversion.
Main Methods:
- Developed an APSL algorithm integrated with Backpropagation Neural Network (BPNN) and Hidden Markov Model (HMM).
- Implemented a multi-stage ASR pipeline including noise reduction, speech-pause detection, and feature extraction.
- Conducted comparative evaluations using custom and benchmark datasets (BNC, ANC, COCA, Buckeye, Emu).
Main Results:
- The APSL-BPNN-HMM model achieved 95.7% recall, 92.95% precision, and 94.53% F-score.
- Demonstrated significantly lower Word Error Rate (WER) compared to baseline HMM and BPNN systems.
- Achieved 96% overall accuracy, validating the effectiveness of adaptive learning in ASR.
Conclusions:
- Adaptive learning within probabilistic frameworks enhances ASR robustness and accuracy.
- The proposed APSL-BPNN-HMM model offers a significant advancement in speech recognition technology.
- This approach effectively addresses challenges like accent variability and noise in ASR systems.
