Frequency band limited musical signals objective and subjective quality assessment

Abstract

In this paper, analysis of features of some objective indicators of the audio signals quality, such as the segmental signal-to-noise ratio, log-spectral distortions and bark-spectral distortions, for estimating the quality of musical signals with a limited frequency band has been presented. It has been shown that the interpolation of the test signals allows increasing the reliability of the evaluation using segmental signal-to-noise ratio and log-spectral distortions indices. Results of objective and subjective quality estimation of speech signals distorted due to bandwidth limitation are compared in the earlier work by means of correlation coefficients calculation and construction of correspondence maps. It was assumed that speech signals occupy a frequency band not exceeding 11 kHz. Meanwhile for high-quality musical signal reproduction, a much wider range of frequencies up to 20 kHz is required. In this regard, it is interesting to study the relevance of subjective and objective indicators of quality for musical signals, which is the main purpose of this work. Experimental studies used 16 second fragments of musical compositions, categories "pop" and "classical" music. The file format is WAVE with a sampling rate of 44.1 kHz. Increasing of the sampling frequency by means of their interpolation has a beneficial effect on segmental signal-to-noise ratio estimates allowing to eliminate illogical violations of the monotonicity of the segmental signal-to-noise ratio dependence on the frequency bandwidth Studies have shown that one of the reasons for the monotonicity violation in the log-spectral distortions dependence is the imperfect frequency response rectangularity of used low-pass filters. Indeed, the log-spectral distortions dependences are monotonic when filtering is realized in the frequency domain by means of ideal rectangular frequency response. It should be noted that when using low-pass filters with rectangular frequency response, the log-spectral distortions value also turned out to be dependent on sampling frequency selection as in the case of non-ideal low-pass filters. The reasons for this phenomenon are worth exploring in the future. Although 30 listeners took part in subjective trials, this quantity can hardly be considered as representative, and further its volume should be increased. Note in addition that quantity of the tested music fragments was very limited. Taking into account the pointed features, the results obtained in this paper should be regarded as preliminary. Obtained results of objective evaluation are compared with the quality assessments obtained by subjective testing. Ref. 9, fig. 3.

Keywords

References

1. Y. O. Mytiai, K. S. Zamsha, B. V. Lozinskyi, O. S. Stepanovska, and A. M. Prodeus, “Objective and subjective assessment of bandlimited signaling speech quality,” Electron. Commun., vol. 21, no. 1, pp. 18–26, Nov. 2016. DOI: 10.20535/2312-1807.2016.21.1.82250
2. A. N. Prodeus, V. S. Didkovskyi, and I. V. Kotvytskyi, “Pro otsinku yakosi movlennia z vykorystanniam indikatoriv loharifmichnykh spektralnykh spotvoren [On the estimation of speech quality using the indicator logarithmic spectral distortions],” in 5th International Scientific and Technical Conference Information Problems in the Theory of Acoustic, Radioelectronic and Telecommunication Systems, IPST-2016, 2016, p. 14.
3. A. Prodeus, “On Some Features of Log-Spectral Distortion as Speech Quality Measure,” Autom. Softw. Dev. Eng. J., vol. 1, 2016. URL: https://asdej.xyz/asde-journal-vol-1/
4. N. V. Bogdanova and A. M. Prodeus, “Objective quality evaluation of speech band-limited signals,” Electron. Commun., vol. 19, no. 6, pp. 58–65, Dec. 2014. DOI: 10.20535/2312-1807.2014.19.6.113479
5. A. Prodeus, “Reducing Sensitivity of Segmental Signal-to-Noise Ratio Estimator to Time-Alignment Error,” Int. J. Electr. Electron. Sci., vol. 2, no. 2, pp. 31–36, 2015. URL: http://www.aascit.org/journal/archive2?journalId=915&paperId=2822
6. S. R. Quackenbush, T. P. Barnwell, and M. A. Clements, Objective measures of speech quality. Englewood Cliffs, N.J.: Prentice Hall, 1988. ISBN: 978-0136290568
7. N. V. Bogdanova and A. M. Prodeus, “Otsinka vplivu neliniinosti fazovoi chastotnoi kharakterystyky traktu na yakist muzychnykh sihnaliv [Assessment of the impact of system phase response non-linearity on the music signals quality],” Electron. Commun., vol. 20, no. 4, pp. 29–35, 2015. DOI: 10.20535/2312-1807.2015.20.4.69901
8. N. Côté, Integral and Diagnostic Intrusive Prediction of Speech Quality. Springer-Verlag Berlin Heidelberg, 2011. ISBN: 978-3-642-18462-8
9. “Matlab Auditory Toolbox (version 2.0).” [Online]. Available: https://www.mathworks.com/matlabcentral/newsreader/view_thread/8417.

License

Copyright (c) 2018 Ігор Валерійович Котвіцький, Андрій Оса (Автор)