Signal to Distortion Ratio

Signal-to-Distortion Ratio (SDR) is a key metric for evaluating the quality of separated audio signals in various applications, including music source separation, speech enhancement, and language-queried audio separation. Current research focuses on improving SDR in real-time applications, often employing lightweight neural network architectures like MMDenseNet and Hybrid Spectrogram-TasNet, and addressing challenges like processing distortions and the need for reference signals. These advancements are crucial for improving the performance of audio processing systems in diverse fields, ranging from hearing aids and live music mixing to robust speech recognition in noisy environments.