Context-aware data augmentation for enhanced speech command recognition in industrial environments.

Area of Science:

• Human-Robot Interaction
• Industrial Automation
• Speech Technology

Background:

• Speech is a key communication channel in Human-Robot Interaction (HRI).
• Industry 4.0 environments require efficient and accurate speech command recognition.
• Challenges include balancing command accuracy with noise and irrelevant speech rejection.

Purpose of the Study:

• To propose a modular framework for optimizing speech command recognition accuracy and robustness in industrial settings.
• To minimize the need for extensive industrial datasets.
• To enhance reliability in noisy production line environments.

Main Methods:

• Developed an efficient Command Recognition module using laboratory data augmented with synthetic samples.
• Implemented context-aware data augmentation and dynamic noise injection for robustness.
• Integrated a Keyword Spotting module to activate recognition upon detecting a predefined keyword.

Main Results:

• The system demonstrated high recall rates for both command recognition and noise rejection.
• Evaluated using real-world samples from a noisy industrial setting.
• Confirmed effectiveness in demanding industrial applications.

Conclusions:

• The proposed modular framework effectively addresses the challenges of speech command recognition in Industry 4.0.
• The system achieves high performance in accuracy and noise rejection.
• This approach enhances productivity and efficiency on industrial production lines.