Both auditory and visual perception serve as two primary inputs through which humans experience and interact with the world. While visual technologies have made significant advancements in recent years, the field of intelligent audio processing continues to face substantial challenges, particularly in environments where noise interference and device limitations severely hinder the accuracy and range of sound capture.
Technological Breakthroughs of Oleap
Oleap introduces an innovative solution by integrating microphone arrays, acoustic scene analysis (ASA), deep learning, and Gammatone filter bank technologies. This comprehensive system not only excels at noise suppression, echo cancellation, and array gain but also offers advanced capabilities in separating, tracking, and enhancing target sound signals, dramatically improving signal quality and providing an essential foundation for subsequent processing.
Near-field environments: Sound enhancement performance can exceed 40dB.
Mid-to-far-field environments: Processing distances can extend beyond 10 meters, with sound enhancement performance improved by 25dB.
Key Innovations and Core Modules
1.Microphone Array Technology: By employing two or more microphones arranged in an array, and leveraging beamforming algorithms, sound signals from various directions can be selectively amplified. This technology effectively suppresses noise from specific directions, and the enhancement effect is contingent on the layout, number, and spatial distribution of the microphone array.
2.Acoustic Scene Analysis (ASA): Mimicking the human ear's auditory capabilities, ASA technology analyzes the energy, phase and other high dimensional spaces relationship between target sound sources and background noise through time-frequency unit decomposition, ensuring noise suppression while enhancing sound signals. Oleap's intelligent sound front-end processing system adapts to different application scenarios by decomposing sound signals into 16 to 256 sub-bands, each processed with individualized intelligent algorithms, achieving an optimal balance between processing efficiency and responsiveness.
3.Deep Learning and Gammatone Filter Bank: By training on vast datasets of noisy sound signals and target sound signals, the system applies deep learning to simulate the biological characteristics of the human ear, enabling precise feature extraction and target sound identification in a variety of noise environments.
Oleap's intelligent sound front-end processing technology has successfully overcome many long-standing challenges in this field, especially in highly challenging application scenarios.:
1.Noise Reduction in Extreme Noise Conditions with Signal Fidelity: In environments with strong noise such as aircraft engines, where noise levels can reach up to 140dB, the noise reduction performance must achieve more than 40dB. However, traditional noise reduction algorithms often lead to significant distortion of the target sound signal, sometimes rendering the signal unrecoverable. Therefore, intelligent audio processing in such environments must not only achieve the required noise reduction levels but also ensure minimal distortion of the target sound, maintaining a MOS score of above 3.5.
2.Balancing Distance of Sound Capture with Signal Quality: Since sound energy inversely correlates with the square of the propagation distance, capturing sound signals at greater distances results in a weakening of the target signal while amplifying the effects of surrounding noise. In outdoor environments where sound capture needs to occur over distances exceeding 10 meters, traditional directional pickup techniques and acoustic design approaches fail to maintain signal quality over such long ranges.
3. Limitations in mono Sound Signal Enhancement: In many real-world scenarios, sound signals are captured through a single microphone. Currently, conventional signal processing techniques face significant limitations in enhancing mono sound signal, typically performing well only with steady-state noise while struggling with more complex noise patterns or signals that share similar spectral characteristics with the target sound.
4.System Complexity and Cost Constraints: To achieve higher sound processing performance, there is often a need to increase the number of microphones, expand CASA (Computational Auditory Scene Analysis) channels, and enhance the complexity of deep learning algorithms. This leads to a rapid increase in system complexity, resulting in higher demands on hardware resources, power consumption, and processing latency, which in many cases exceeds the cost and resource limitations of practical applications, thus hindering the broader adoption of such technologies.
