Method for learning an audio quality metric combining labeled and unlabeled data
Described is a method of training a neural-network-based system for determining an indication of an audio quality of an audio input. The method includes obtaining, as input, at least one training set comprising audio samples. The audio samples include audio samples of a first type and audio samples of a second type, wherein each of the first type of audio samples is labelled with information indicative of a respective predetermined audio quality metric, and wherein each of the second type of audio samples is labelled with information indicative of a respective audio quality metric relative to that of a reference audio sample. The method further includes: inputting the training set to the neural-network-based system; and iteratively training the system to predict the respective label information of the audio samples in the training set.
1 . A method of training a neural-network-based system for determining an indication of an audio quality of an audio input, the method comprising:
obtaining, as input, at least one training set comprising audio samples, wherein the audio samples comprise audio samples of a first type and audio samples of a second type, wherein each of the first type of audio samples is labelled with information indicative of a respective predetermined audio quality metric, and wherein each of the second type of audio samples is labelled with information indicative of a respective audio quality metric relative to that of a reference audio sample;
inputting the audio samples of the training set to the neural-network-based system including:
a neural-network-based encoder configured to map an audio sample to a respective latent vector in a latent space, and
a neural-network-based assessment stage including a first assessment head and one or more second assessment heads, the first assessment head being configured to generate a respective quality score corresponding to the audio sample based on the respective latent vector; and
iteratively training the neural-network-based system to predict the respective label information of the audio samples in the training set based on a plurality of loss functions,
wherein each of the loss functions is configured to reflect differences between the respective label information of the audio samples in the training set and the respective predictions thereof generated by a respective one of the assessment heads; and
wherein the one or more second assessment heads and a corresponding subset of the plurality of loss functions are configured to regularize the latent space based on a plurality of respective latent vectors to which the audio samples of the training set are mapped during the iterative training.
2 . The method according to claim 1 , wherein the first type of audio samples comprises human annotated audio samples, each being labelled with the information indicative of the respective predetermined audio quality metric, wherein the human annotated audio samples comprise mean opinion score, MOS, audio samples and/or just-noticeable difference, JND, audio samples.
3 . The method according to claim 2 , wherein the second type of audio samples comprises algorithmically generated audio samples, each being labelled with the information indicative of the relative audio quality metric.
4 . The method according to claim 3 ,
wherein each of the algorithmically generated samples is generated by selectively applying at least one degradation function, each with a respective degradation strength, to a reference audio sample or to another algorithmically generated audio sample, and
wherein the label information comprises information indicating the respective degradation function and/or the respective degradation strength that have been applied thereto.
5 . The method according to claim 4 , wherein the label information further comprises information indicative of degradation relative to the reference audio sample or to the other audio sample in the training set, wherein the degradation function is selected from a plurality of available degradation functions, and/or wherein the respective degradation strength is set such that, at its minimum, the degradation is perceptually noticeable.
6 . The method according to claim 5 , wherein the plurality of available degradation functions comprises functions relating to one or more of: reverberation, clipping, encoding with different codecs, phase distortion, audio reversing, and background noise.
7 . The method according to claim 4 , wherein the algorithmically generated audio samples are generated as pairs of audio frames {x i , x j } and/or quadruples of audio frames {x i k , x i l , x j k , x j l }, wherein the audio frame x i is generated by selectively applying at least one degradation function each with a respective degradation strength to a reference audio frame, wherein the audio frame x j is generated by selectively applying at least one degradation function each with a respective degradation strength to the audio frame x i , wherein the audio frames x i k and x i l are extracted from audio frame x i by selectively applying a respective time delay to the audio frame x i , and wherein the audio frames x j k and x j l are extracted from audio frame x j by selectively applying a respective time delay to the audio frame x j .
8 . The method according to claim 3 , wherein the label information of the second type of audio samples comprises relative information indicative of whether one audio sample is more degraded than another audio sample, wherein the plurality of loss functions comprises a loss function indicative of a pairwise ranking metric and calculated based on a ranking established by the label information comprising the relative degradation information and the prediction thereof.
9 . The method according to claim 3 , wherein the label information of the second type of audio samples comprises relative information indicative of perceptual relevance between audio samples, wherein the plurality of loss functions comprises a loss function indicative of a consistency metric, wherein the consistency metric indicates whether two or more audio samples have the same degradation function and degradation strength, and correspond to the same time frame, and wherein the loss function is calculated based on the difference between the label information comprising the perceptual relevance information and the prediction thereof.
10 . The method according to claim 3 , wherein the label information of the second type of audio samples comprises relative information indicative of whether one audio sample has been applied with the same degradation function and the same degradation strength as another audio sample, wherein the plurality of loss functions comprises a loss function indicative of a degradation condition metric, and wherein the loss function is calculated based on the difference between the label information comprising the relative degradation information and the prediction thereof.
11 . The method according to claim 3 , wherein the label information of the second type of audio samples comprises relative information indicative of perceptual difference relative to one another, wherein the plurality of loss functions comprises a loss function indicative of a JND metric, and wherein the loss function is calculated based on the difference between the label information comprising the relative perceptual difference and the prediction thereof.
12 . The method according to claim 3 , wherein the label information of the second type of audio samples comprises information indicative of the degradation function that has been applied to the audio sample, wherein the plurality of loss functions comprises a loss function indicative of a degradation type metric, and wherein the loss function is calculated based on difference between the label information comprising the respective degradation function information and the prediction thereof.
13 . The method according to claim 3 , wherein the label information of the second type of audio samples comprises information indicative of the degradation strength that has been applied to an audio sample, wherein the plurality of loss functions comprises a loss function indicative of a degradation strength metric, and wherein the loss function is calculated based on difference between the label information comprising the respective degradation strength information and the prediction thereof.
14 . The method according to claim 3 , wherein the plurality of loss functions comprises a loss function indicative of a regression metric, and wherein the regression metric is calculated according to at least one of reference-based and/or reference-free quality measures, wherein the reference-based quality measures comprise at least one of: PESQ, CSIG, CBAK, COVL, SSNR, LLR, WSSD, STOI, SISDR, Mel cepstral distortion, and log-Mel-band distortion.
15 . The method according to claim 1 , wherein each of the assessment heads comprises a respective linear layer or a respective multilayer perceptron, MLP.
16 . A neural-network-based system for determining an indication of an audio quality of an input audio sample, wherein the neural-network-based system comprises:
a neural-network-based encoder configured to map the input audio sample to a respective latent vector in a latent space; and
a neural-network-based assessment stage including a first assessment head and one or more second assessment heads,
wherein the first assessment head is configured to generate a respective quality score corresponding to the input audio sample based on the respective latent vector; and
wherein the one or more second assessment heads are configured to regularize the latent space based on a plurality of latent vectors to which audio samples of a training set are mapped by the neural-network-based encoder during iterative training of the neural-network-based system.
17 . A method of operating a neural-network-based system for determining an indication of an audio quality of an input audio sample, wherein the system comprises a neural-network-based encoder and a neural-network-based assessment stage including a plurality of assessment heads, the method comprising:
mapping, by the neural-network-based encoder, the input audio sample to a respective latent vector in a latent space; and
predicting for the input audio sample,
by a first subset of the plurality of assessment heads of the neural-network-based assessment stage, information indicative of a predetermined audio quality metric, and
by a second subset of the plurality of assessment heads of the neural-network-based assessment stage, information indicative of a relative audio quality metric relative to a reference audio sample,
wherein the predicting by the first and second subsets of the plurality of assessment heads is based on the respective latent vector.