Single-Ended Speech Quality Measurement Using Machine Learning Methods

Abstract

Abstract-We describe a novel single-ended algorithm constructed from models of speech signals, including clean and degraded speech, and speech corrupted by multiplicative noise and temporal discontinuities. Machine learning methods are used to design the models, including Gaussian mixture models, support vector machines, and random forest classifiers. Estimates of the subjective mean opinion score (MOS) generated by the models are combined using hard or soft decisions generated by a classifier which has learned to match the input signal with the models. Test results show the algorithm outperforming ITU-T P.563, the current "state-of-art" standard single-ended algorithm. Employed in a distributed double-ended measurement configuration, the proposed algorithm is found to be more effective than P.563 in assessing the quality of noise reduction systems and can provide a functionality not available with P.862 PESQ, the current double-ended standard algorithm.Index Terms-Mean opinion score (MOS), objective quality measurement, quality model, single-ended measurement, speech communication, speech distortions, speech enhancement, speech quality, subjective quality.

Hybrid Signal-and-Link-Parametric Speech Quality Measurement for VoIP Communications

“…The auditory spectrum is approximated by an all-pole autoregressive model, whose coefficients are transformed to th-order PLP cepstral coefficients . The zeroth cepstral coefficient is employed as an energy measure [49], and is chosen from previous experiments [50]. When describing the PLP vector for a given frame , the notation and will be used.…”

“…On the other hand, approximately 10% of "normal" abrupt starts, such as those experienced with certain plosive consonants (e.g.,/d/), are misclassified as clippings. To improve classification performance, more complex machine learning methods can be used [50]. Since abrupt starts have, intuitively, less significant impact on perceived speech quality [30], [56], such classification errors are shown not to be detrimental to overall speech quality measurement.…”

A Hybrid Signal-and-Link-Parametric Approach to Single-Ended Quality Measurement of Packetized Speech

“…The zeroth cepstral coefficient is used as a log-energy term. We also experiment with delta and double-delta coefficients [11] as measures of signal spectral dynamics.…”

Effectiveness of ideal ratio mask for non-intrusive quality assessment of noise suppressed speech

“…Different features extracted from speech have been detected to be useful for speech quality assessment. Spectral dynamics, spectral flatness, spectral centroid, spectral variance, fundamental frequency or pitch (F 0 ) excitation variance and perceptual linear prediction (PLP) coefficients were used for quality prediction in [5], [6]. In [7] and [8], the quality assessment problem is posed as a regression problem and the mapping between acoustic features and the subjective score was found using Mel Frequency Cepstral Coefficients (MFCCs) and filterbank energies, respectively.…”