Enhancing human computer interaction with coot optimization and deep learning for multi language identification

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Human-Computer Interaction (HCI) focuses on designing effective and user-friendly computer systems.
• Multiple Spoken Language Identification (MSLI) is crucial for seamless HCI, enabling systems to recognize diverse languages.
• Deep Learning (DL) techniques, particularly neural networks, have shown great promise in speech and language processing tasks.

Purpose of the Study:

• To develop a novel technique for Multiple Spoken Language Identification (MSLI) within Human-Computer Interaction (HCI) applications.
• To enhance the accuracy and robustness of language detection systems, regardless of speaker characteristics.
• To introduce the Coot Optimizer Algorithm with DL-Driven Multiple SLI and Detection (COADL-MSLID) for advanced HCI.

Main Methods:

• Audio files are converted into spectrogram images for analysis.
• The SqueezeNet model is utilized for feature vector extraction.
• The Coot Optimizer Algorithm (COA) optimizes hyperparameters for SqueezeNet, and a Convolutional Autoencoder (CAE) is employed for Spoken Language Identification (SLID).

Main Results:

• The proposed COADL-MSLID technique demonstrates high performance in detecting multiple spoken languages.
• Experimental validation on a benchmark dataset yielded an accuracy of 98.33%.
• The method effectively identifies languages across different genders, speaking styles, and ages.

Conclusions:

• The COADL-MSLID technique offers a significant advancement in MSLI for HCI applications.
• The integration of COA and DL models provides a robust and accurate solution for spoken language detection.
• This approach enhances the naturalness and efficiency of human-computer interactions through improved language recognition.