Next: Reference Segmentation Selection and Up: Evaluation Metrics Previous: Diarization Error Rate Contents
The SNR is a metric based on the power ratio between the signal and the noise. It is mainly used in signal processing applications to evaluate how the desired signal stands out from the background noise. It this thesis it was thought useful to measure how much quality does the resulting speech signal has after the beamforming module. In speech signals it is not clear what part belongs to silence and what to speech, therefore several methods can be applied to compute an SNR approximation. Each computation method can be very independent from every other and therefore comparisons should only be made using the same estimation algorithm. In this thesis the method used in the NIST Speech Quality Assurance Package (SPQA) (NIST Speech tools and APIs, 2006) and described in detail in section 3.2.1.