This paper presents NOMAD (Non-Matching Audio Distance), a differentiable perceptual similarity metric that measures the distance of a degraded signal against non-matching references. The proposed method is based on learning deep feature embeddings via a triplet loss guided by the Neurogram Similarity Index Measure (NSIM) to capture degradation intensity. During inference, the similarity score between any two audio samples is computed through Euclidean distance of their embeddings. NOMAD is fully unsupervised and can be used in general perceptual audio tasks for audio analysis e.g. quality assessment and generative tasks such as speech enhancement and speech synthesis. The proposed method is evaluated with 3 tasks. Ranking degradation intensity, predicting speech quality, and as a loss function for speech enhancement. Results indicate NOMAD outperforms other non-matching reference approaches in both ranking degradation intensity and quality assessment, exhibiting competitive performance with full-reference audio metrics. NOMAD demonstrates a promising technique that mimics human capabilities in assessing audio quality with non-matching references to learn perceptual embeddings without the need for human-generated labels.
Self-supervised representation learning (SSRL) has improved the performance on downstream phoneme recognition versus supervised models. Training SSRL models requires a large amount of pre-training data and this poses a challenge for low resource languages. A common approach is transferring knowledge from other languages. Instead, we propose to use audio augmentation to pre-train SSRL models in a low resource condition and evaluate phoneme recognition as downstream task. We performed a systematic comparison of augmentation techniques, namely: pitch variation, noise addition, accented target-language speech and other language speech. We found combined augmentations (noise/pitch) was the best augmentation strategy outperforming accent and language knowledge transfer. We compared the performance with various quantities and types of pre-training data. We examined the scaling factor of augmented data to achieve equivalent performance to models pre-trained with target domain speech. Our findings suggest that for resource constrained languages, in-domain synthetic augmentation can outperform knowledge transfer from accented or other language speech.
The development of data-driven heart sound classification models has been an active area of research in recent years. To develop such data-driven models in the first place, heart sound signals need to be captured using a signal acquisition device. However, it is almost impossible to capture noise-free heart sound signals due to the presence of internal and external noises in most situations. Such noises and degradations in heart sound signals can potentially reduce the accuracy of data-driven classification models. Although different techniques have been proposed in the literature to address the noise issue, how and to what extent different noise and degradations in heart sound signals impact the accuracy of data-driven classification models remains unexplored. To answer this question, we produced a synthetic heart sound dataset including normal and abnormal heart sounds contaminated with a large variety of noise and degradations. We used this dataset to investigate the impact of noise and degradation in heart sound recordings on the performance of different classification models. The results show different noises and degradations affect the performance of heart sound classification models to a different extent; some are more problematic for classification models, and others are less destructive. Comparing the findings of this study with the results of a survey we previously carried out with a group of clinicians shows noise and degradations that are more detrimental to classification models are also more disruptive to accurate auscultation. The findings of this study can be leveraged to develop targeted heart sound quality enhancement approaches - which adapt the type and aggressiveness of quality enhancement based on the characteristics of noise and degradation in heart sound signals.
Learning music representations that are general-purpose offers the flexibility to finetune several downstream tasks using smaller datasets. The wav2vec 2.0 speech representation model showed promising results in many downstream speech tasks, but has been less effective when adapted to music. In this paper, we evaluate whether pre-training wav2vec 2.0 directly on music data can be a better solution instead of finetuning the speech model. We illustrate that when pre-training on music data, the discrete latent representations are able to encode the semantic meaning of musical concepts such as pitch and instrument. Our results show that finetuning wav2vec 2.0 pre-trained on music data allows us to achieve promising results on music classification tasks that are competitive with prior work on audio representations. In addition, the results are superior to the pre-trained model on speech embeddings, demonstrating that wav2vec 2.0 pre-trained on music data can be a promising music representation model.
Non-reference speech quality models are important for a growing number of applications. The VoiceMOS 2022 challenge provided a dataset of synthetic voice conversion and text-to-speech samples with subjective labels. This study looks at the amount of variance that can be explained in subjective ratings of speech quality from metadata and the distribution imbalances of the dataset. Speech quality models were constructed using wav2vec 2.0 with additional metadata features that included rater groups and system identifiers and obtained competitive metrics including a Spearman rank correlation coefficient (SRCC) of 0.934 and MSE of 0.088 at the system-level, and 0.877 and 0.198 at the utterance-level. Using data and metadata that the test restricted or blinded further improved the metrics. A metadata analysis showed that the system-level metrics do not represent the model's system-level prediction as a result of the wide variation in the number of utterances used for each system on the validation and test datasets. We conclude that, in general, conditions should have enough utterances in the test set to bound the sample mean error, and be relatively balanced in utterance count between systems, otherwise the utterance-level metrics may be more reliable and interpretable.
This paper introduces a comparison of deep learning-based techniques for the MOS prediction task of synthesised speech in the Interspeech VoiceMOS challenge. Using the data from the main track of the VoiceMOS challenge we explore both existing predictors and propose new ones. We evaluate two groups of models: NISQA-based models and techniques based on fine-tuning the self-supervised learning (SSL) model wav2vec2_base. Our findings show that a simplified version of NISQA with 40% fewer parameters achieves results close to the original NISQA architecture on both utterance-level and system-level performances. Pre-training NISQA with the NISQA corpus improves utterance-level performance but shows no benefit on the system-level performance. Also, the NISQA-based models perform close to LDNet and MOSANet, 2 out of 3 baselines of the challenge. Fine-tuning wav2vec2_base shows superior performance than the NISQA-based models. We explore the mismatch between natural and synthetic speech and discovered that the performance of the SSL model drops consistently when fine-tuned on natural speech samples. We show that adding CNN features with the SSL model does not improve the baseline performance. Finally, we show that the system type has an impact on the predictions of the non-SSL models.
Recent studies have shown how self-supervised models can produce accurate speech quality predictions. Speech representations generated by the pre-trained wav2vec 2.0 model allows constructing robust predicting models using small amounts of annotated data. This opens the possibility of developing strong models in scenarios where labelled data is scarce. It is known that fine-tuning improves the model's performance; however, it is unclear how the data (e.g., language, amount of samples) used for fine-tuning is influencing that performance. In this paper, we explore how using different speech corpus to fine-tune the wav2vec 2.0 can influence its performance. We took four speech datasets containing degradations found in common conferencing applications and fine-tuned wav2vec 2.0 targeting different languages and data size scenarios. The fine-tuned models were tested across all four conferencing datasets plus an additional dataset containing synthetic speech and they were compared against three external baseline models. Results showed that fine-tuned models were able to compete with baseline models. Larger fine-tune data guarantee better performance; meanwhile, diversity in language helped the models deal with specific languages. Further research is needed to evaluate other wav2vec 2.0 models pre-trained with multi-lingual datasets and to develop prediction models that are more resilient to language diversity.
Audio quality assessment has been widely researched in the signal processing area. Full-reference objective metrics (e.g., POLQA, ViSQOL) have been developed to estimate the audio quality relying only on human rating experiments. To evaluate the audio quality of novel audio processing techniques, researchers constantly need to compare objective quality metrics. Testing different implementations of the same metric and evaluating new datasets are fundamental and ongoing iterative activities. In this paper, we present AQP - an open-source, node-based, light-weight Python pipeline for audio quality assessment. AQP allows researchers to test and compare objective quality metrics helping to improve robustness, reproducibility and development speed. We introduce the platform, explain the motivations, and illustrate with examples how, using AQP, objective quality metrics can be (i) compared and benchmarked; (ii) prototyped and adapted in a modular fashion; (iii) visualised and checked for errors. The code has been shared on GitHub to encourage adoption and contributions from the community.
Non-intrusive speech quality assessment is a crucial operation in multimedia applications. The scarcity of annotated data and the lack of a reference signal represent some of the main challenges for designing efficient quality assessment metrics. In this paper, we propose two multi-task models to tackle the problems above. In the first model, we first learn a feature representation with a degradation classifier on a large dataset. Then we perform MOS prediction and degradation classification simultaneously on a small dataset annotated with MOS. In the second approach, the initial stage consists of learning features with a deep clustering-based unsupervised feature representation on the large dataset. Next, we perform MOS prediction and cluster label classification simultaneously on a small dataset. The results show that the deep clustering-based model outperforms the degradation classifier-based model and the 3 baselines (autoencoder features, P.563, and SRMRnorm) on TCD-VoIP. This paper indicates that multi-task learning combined with feature representations from unlabelled data is a promising approach to deal with the lack of large MOS annotated datasets.
Good speech quality has been achieved using waveform matching and parametric reconstruction coders. Recently developed very low bit rate generative codecs can reconstruct high quality wideband speech with bit streams less than 3 kb/s. These codecs use a DNN with parametric input to synthesise high quality speech outputs. Existing objective speech quality models (e.g., POLQA, ViSQOL) do not accurately predict the quality of coded speech from these generative models underestimating quality due to signal differences not highlighted in subjective listening tests. We present WARP-Q, a full-reference objective speech quality metric that uses dynamic time warping cost for MFCC speech representations. It is robust to small perceptual signal changes. Evaluation using waveform matching, parametric and generative neural vocoder based codecs as well as channel and environmental noise shows that WARP-Q has better correlation and codec quality ranking for novel codecs compared to traditional metrics in addition to versatility for general quality assessment scenarios.