The Tacotron-Based Signal Synthesis Method for Active Sonar.

Area of Science:

• Acoustics
• Artificial Intelligence
• Signal Processing

Background:

• Active sonar systems face challenges in detecting underwater targets due to submarine quieting and increased maritime traffic.
• Multipath propagation and low signal-to-noise ratios complicate target detection, tracking, and identification.
• Existing machine learning methods require extensive data, while mathematical modeling struggles with complex underwater environments.

Purpose of the Study:

• To propose an artificial intelligence-based technique for synthesizing active sonar signals.
• To overcome the limitations of data scarcity and inaccurate modeling in current sonar signal generation.
• To adapt the Tacotron speech synthesis model for generating realistic active sonar data.

Main Methods:

• Modified the Tacotron model, a speech synthesis architecture, for sonar signal generation.
• Applied artificial intelligence to synthesize active sonar data, addressing limitations of mathematical modeling.
• Utilized spectrogram analysis and mean opinion score evaluation to validate the synthesized signals.

Main Results:

• The proposed method successfully synthesized active sonar data.
• Spectrogram analysis confirmed the similarity of synthesized data to trained data.
• Mean opinion scores indicated the high quality and realism of the generated sonar signals.

Conclusions:

• The AI-based sonar signal synthesis technique effectively generates active sonar data comparable to real-world data.
• This method offers a viable solution for augmenting datasets in active sonar research.
• The adapted Tacotron model shows promise for advancing sonar signal processing and underwater acoustics.