Evaluation Metrics

1. Speech Intelligibility

1.1 Signal-to-Distortion Ratio (SDR)

Signal-to-Distortion Ratio (SDR) is an objective evaluation metric used to measure the quality of reconstructed signals in speech separation tasks. SDR assesses the level of distortion by calculating the error between the reconstructed signal and the reference signal. A higher SDR value indicates better performance in speech separation and higher quality of the reconstructed signal.

1.2 Scale-Invariant Signal-to-Distortion Ratio (SI-SDR)

Scale-Invariant Signal-to-Distortion Ratio (SI-SDR) is a commonly used objective evaluation metric for assessing the quality of reconstructed signals in speech separation tasks. SI-SDR evaluates the level of distortion by comparing the energy ratio between the enhanced speech and the reference speech while ignoring their amplitude differences. The calculation formula of SI-SDR eliminates the influence of signal amplitude, thus focusing more on the shape and content of the signal. A higher SI-SDR value indicates better performance in speech separation and higher quality of the reconstructed signal.