Single and multi-channel speech enhancement and audio scene analysis algorithms targeting IoT applications.
• Designed and implemented algorithms mixing deep learning and conventional DSP for audio scene classification, noise/echo suppression and signal conditioning for voice commands (barge-in and ASR).
• Built machine learning training framework using Python and TensorFlow (feature extraction, label creation, neural network training and testing).
• Built exhaustive audio data sets to create realistic audio scenes for training and testing of machine learning algorithms.
• Introduced objective quality metrics for human and machine listeners.
• Developed data-driven noise suppression algorithms for multi-microphone systems for close and far-field mobile communications.