-1
$\begingroup$

I need an algorithm to assess the quality of a given audio. The task, is considered to be no-reference. Can anyone help me whether there is any method for quality assessment of any given audio?

edit1: Note that I do not want to just classify them to good/bad (or similarly high/low quality). I need to give a score(similar to MOS) to show how much good/bad is the given audio based on human auditory system (human perception).

edit2: I need to describe the above question more, because some people have marked it as to be duplicate. See, I need to predict the subjective score. This question was not asked at StackExcahnge, before. In Analyzing the quality of a music track a different question has been asked. As its author (@robert-lemiesz) has explained in the 3rd comment of the main post, he was not looking for predicting subjective score. But, I am looking for predicting subjective score. Please, remove your vote from this post of being duplicate and kindly help me to get to the correct answer.

$\endgroup$
  • 1
    $\begingroup$ basically, there's no such thing as "objective" audio quality. Is a crystal-clean recording of some synthesizer-driven pop music "better" in quality than the muddy, distorted original recording of Jimi Hendrix' All Along The Watchtower? $\endgroup$ – Marcus Müller Aug 3 '18 at 13:12
  • $\begingroup$ Can you post two representative examples of two tracks, one which would have to be given the tag "low quality" and one which would have to be given the tag "high quality"? You may be able to train a model given specific characteristics to perform the filtering you are after with some accuracy. $\endgroup$ – A_A Aug 4 '18 at 7:09
  • $\begingroup$ Thanks you guys for your comments. Actually, as I described in the edit of the post, I do not want to just classify the audio into good/bad quality. I need to give a score(similar to MOS) to show how much good/bad is the audio. $\endgroup$ – Quasar Aug 4 '18 at 18:05
  • $\begingroup$ Is it speech or is it music, or something else? $\endgroup$ – Olli Niemitalo Aug 4 '18 at 18:19
  • 1
    $\begingroup$ what is the "quality" being assessed? what does "high quality" or "low quality" mean? how can you describe this meaning without a reference signal? $\endgroup$ – robert bristow-johnson Aug 5 '18 at 3:36
3
$\begingroup$

Audio quality assessment is one of the most critical pieces of audio coding and enhancing applications. The task requires an accurate and objective (mathematical) modeling of human auditory system including its subjective virtues.

However, the task of subjective quality assessment is one of the most complex problems to be attacked on Earth. Currently all available models mostly deal with basic things like quantization, SNR, bandwidth, distortion, masking, delay, echo, frequency response, etc... Achievements are like, detecting speech or music, recognition of musical instruments and acoustical environments, determination of the musical genre etc.

Theoretical studies for modeling an accurate human auditory system are carried out at audio signal processing laboratoaries around the world for designing automated audio compression algorithms to optimize the bitrate of audio based on the computed (percepted) audio quality.

$\endgroup$
  • $\begingroup$ Thanks @fat32 . So, you mean, so far, there is no way to assess the quality of any given audio in a way that its result would have high correlation coefficient with human opinion score? $\endgroup$ – Quasar Aug 4 '18 at 20:07
  • $\begingroup$ @Quasar I don't say there is no way; there may be some but exremely complex. The issue subjectivity is beyond the reach of compemporary science and engineering... I hope this suffices to describe the seriousness of the problem ;-) $\endgroup$ – Fat32 Aug 4 '18 at 20:19
  • $\begingroup$ Dear @fat32 , I understand how complex it is. But, I was wondering if there has been any way proposed for this issue, yet? $\endgroup$ – Quasar Aug 5 '18 at 17:07
  • $\begingroup$ @Quasar you can find such a model from Fraunhofer Institute. Send them a mail for asking an evaluation copy of the human hearing system model... $\endgroup$ – Fat32 Aug 5 '18 at 22:19
1
$\begingroup$

By quality, one might mean intelligibility, which may be reduced by noise and other signal degradation. This kind of assumes that there is a reference signal that gets degraded, although the reference signal is not available.

One idea is to measure Gaussianity of the signal. From the abstract of Ganesh R. Naik & Dinesh K. Kumar (2011) Improving the quality of the audio sources using Gaussianity reduction technique, International Journal of Electronics, 98:7, 949-959:

Background signals are a result of the inferential summation of large number of sources, while the foreground signals are a result of limited number of sources. This makes the statistical properties of the signal very different.

This is related to summing of independent signals and the central limit theorem. Likewise, reverb and echos make the audio less intelligible, and those also make the signal's distribution more Gaussian if the original signal is not correlated with itself much over such long time lags.

Distortion (signal clipping) will reduce Gaussianity, which may be a problem with this approach. Then again distortion may actually improve (speech) intelligibility.

$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.