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

Abstract

A hybrid signal-and-link-parametric approach to single-ended quality measurement of packetized speech is proposed. Transmission link parameters are used to determine a base quality for the test signal. The base quality is adjusted by degradation factors calculated from perceptual features extracted from the test signal. The degradation factors are based on KullbackLeibler distances between a parametric model trained online for the extracted features and reference models of normative speech behavior. The proposed method overcomes the limitations of pure link parametric and pure signal-based methods.

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

“…ITU-T Recommendation P.563 represents the current state-of-art single-ended algorithm for traditional telephony applications [31]. Recent research, however, has suggested that P.563 performance is also compromised for VoIP applications [6]- [8]. In addition, the experiments described in Section III also suggest algorithm sensitivity to different VoIP connection parameters and high per-call estimation errors.…”

“…Large may result in a model that overfits the training data, whereas small may result in models that are not accurate. Commonly, the approach taken is to try several values of and use the one that results in superior performance on an unseen test set [8]. Here, an alternate approach is taken and a recursive (greedy) EM algorithm is used.…”

“…In [8], signal-based distortions were measured by means of the Kullback-Leibler distance between offline-trained reference models and GM models obtained online from the test speech signal. While the approach was shown to be effective, real-time operation becomes questionable as speech packets need to be stored in order for accurate online GM models to be obtained.…”

“…A hybrid signal-and-link-parametric approach to single-ended quality measurement of packet speech is proposed, based on extending the work described in [8]. The method makes use of IP connection parameters such as codec and PLC type, packet size, and packet loss pattern to determine a "base quality" representative of specific transmission links.…”

“…While signal-based algorithms perform well for traditional telephony applications, recent research has shown that algorithm performance may decrease when applied to VoIP communications. Several independent studies have shown that signal based schemes can be sensitive to connection parameters [5]- [8], as well as attain high "per-call" quality estimation errors [8]- [10]; such limitations restrict the usefulness of signal-based methods for online quality monitoring and control applications. Additionally, high computational complexity also acts as a major limiting factor for the widespread use of standard signal-based measures for online QoS control.…”