Objective and subjective speech quality evaluation of wideband noise reduction algorithms.

Abstract

Subjective and/or objective measurements of speech quality are important in benchmarking speech enhancement algorithms. Subjective measures include ratings of speech quality by listeners, whereas objective measures compute a metric based on the clean and enhanced speech samples. While subjective quality ratings are the “gold-standard,” they are also time- and resource-consuming. An objective metric that correlates highly with subjective data is attractive, as it can act as a substitute for benchmarking and fine-tuning the noise reduction algorithms. In this paper, the performance of several noise reduction algorithms for wideband (50–7000 Hz) telephony application was evaluated both subjectively and objectively. A custom wideband noise reduction database was created that contained speech samples corrupted by different background noises at different signal to noise ratios and processed by seven different noise reduction algorithms. Speech samples in the database were subsequently rated by a group of 32 listeners with normal hearing capabilities. Several objective metrics including log-likelihood ratio, weighted spectral slope, PESQ, and the loudness pattern distortion (LPD) measure based on the Moore–Glasberg auditory model were used to predict the subjective ratings. Results showed that the subjective ratings were highly reliable and the LPD metric correlated the best with subjective ratings of enhanced wideband speech.