Simulation of English Speech Recognition Based on Improved Extreme Random Forest Classification

Area of Science:

• Computer Science
• Artificial Intelligence
• Signal Processing

Background:

• Current speech recognition systems exhibit limitations in handling diverse language types and achieving high recognition rates.
• Research focus is on developing efficient classifiers for improved speech recognition system performance.

Purpose of the Study:

• To develop a high-recognition-rate English speech recognition model using machine learning.
• To enhance speech signal recognition through adaptive denoising and improved classification algorithms.

Main Methods:

• Employed machine learning with a computational random forest classification method for algorithm improvement.
• Utilized a lightweight AlexNet model, incorporating adaptive wavelet threshold shrinkage and denoising on time-frequency images.
• Replaced the softmax function with an EI strong classifier to boost accuracy in low signal-to-noise ratio conditions.

Main Results:

• The developed machine learning model demonstrated effective English speech recognition capabilities.
• Significant improvements in recognition accuracy were observed, especially under low signal-to-noise conditions.
• The adaptive wavelet threshold shrinkage and denoising techniques positively impacted signal processing.

Conclusions:

• The proposed machine learning model offers a promising solution for accurate English speech recognition.
• The integration of random forest and EI strong classifiers enhances robustness in challenging acoustic environments.
• This research contributes to advancing speech recognition technology, particularly for non-mainstream audio types.