Electrical Engineering And Systems Science

In this work, we propose DiffWave, a versatile diffusion probabilistic model for conditional and unconditional waveform generation. The model is non-autoregressive, and converts the white noise signal into structured waveform through a Markov chain with a constant number of steps at synthesis. It is efficiently trained by optimizing a variant of variational bound on the data likelihood. DiffWave produces high-fidelity audios in different waveform generation tasks, including neural vocoding conditioned on mel spectrogram, class-conditional generation, and unconditional generation. We demonstrate that DiffWave matches a strong WaveNet vocoder in terms of speech quality (MOS: 4.44 versus 4.43), while synthesizing orders of magnitude faster. In particular, it significantly outperforms autoregressive and GAN-based waveform models in the challenging unconditional generation task in terms of audio quality and sample diversity from various automatic and human evaluations.

Training convolutional recurrent neural networks on first-order Ambisonics signals is a well-known approach when estimating the direction of arrival for speech/sound signals. In this work, we investigate whether increasing the order of Ambisonics up to the fourth order further improves the estimation performance of convolutional recurrent neural networks. While our results on data based on simulated spatial room impulse responses show that the use of higher Ambisonics orders does have the potential to provide better localization results, no further improvement was shown on data based on real spatial room impulse responses from order two onwards. Rather, it seems to be crucial to extract meaningful features from the raw data. First order features derived from the acoustic intensity vector were superior to pure higher-order magnitude and phase features in almost all scenarios.

Music similarity search is useful for a variety of creative tasks such as replacing one music recording with another recording with a similar "feel", a common task in video editing. For this task, it is typically necessary to define a similarity metric to compare one recording to another. Music similarity, however, is hard to define and depends on multiple simultaneous notions of similarity (i.e. genre, mood, instrument, tempo). While prior work ignore this issue, we embrace this idea and introduce the concept of multidimensional similarity and unify both global and specialized similarity metrics into a single, semantically disentangled multidimensional similarity metric. To do so, we adapt a variant of deep metric learning called conditional similarity networks to the audio domain and extend it using track-based information to control the specificity of our model. We evaluate our method and show that our single, multidimensional model outperforms both specialized similarity spaces and alternative baselines. We also run a user-study and show that our approach is favored by human annotators as well.

Over the recent years, various deep learning-based embedding methods have been proposed and have shown impressive performance in speaker verification. However, as in most of the classical embedding techniques, the deep learning-based methods are known to suffer from severe performance degradation when dealing with speech samples with different conditions (e.g., recording devices, emotional states). In this paper, we propose a novel fully supervised training method for extracting a speaker embedding vector disentangled from the variability caused by the nuisance attributes. The proposed framework was compared with the conventional deep learning-based embedding methods using the RSR2015 and VoxCeleb1 dataset. Experimental results show that the proposed approach can extract speaker embeddings robust to channel and emotional variability.

Speech enhancement techniques based on deep learning have brought significant improvement on speech quality and intelligibility. Nevertheless, a large gain in speech quality measured by objective metrics, such as perceptual evaluation of speech quality (PESQ), does not necessarily lead to improved speech recognition performance due to speech distortion in the enhancement stage. In this paper, a multi-channel dilated convolutional network based frequency domain modeling is presented to enhance target speaker in the far-field, noisy and multi-talker conditions. We study three approaches towards distortionless waveforms for overlapped speech recognition: estimating complex ideal ratio mask with an infinite range, incorporating the fbank loss in a multi-objective learning and finetuning the enhancement model by an acoustic model. Experimental results proved the effectiveness of all three approaches on reducing speech distortions and improving recognition accuracy. Particularly, the jointly tuned enhancement model works very well with other standalone acoustic model on real test data.

The two most common paradigms for end-to-end speech recognition are connectionist temporal classification (CTC) and attention-based encoder-decoder (AED) models. It has been argued that the latter is better suited for learning an implicit language model. We test this hypothesis by measuring temporal context sensitivity and evaluate how the models perform when we constrain the amount of contextual information in the audio input. We find that the AED model is indeed more context sensitive, but that the gap can be closed by adding self-attention to the CTC model. Furthermore, the two models perform similarly when contextual information is constrained. Finally, in contrast to previous research, our results show that the CTC model is highly competitive on WSJ and LibriSpeech without the help of an external language model.

The COVID-19 pandemic has led to a dramatic increase in the use of face masks worldwide. Face coverings can affect both acoustic properties of the signal as well as speech patterns and have unintended effects if the person wearing the mask attempts to use speech processing technologies. In this paper we explore the impact of wearing face masks on the automated assessment of English language proficiency. We use a dataset from a large-scale speaking test for which test-takers were required to wear face masks during the test administration, and we compare it to a matched control sample of test-takers who took the same test before the mask requirements were put in place. We find that the two samples differ across a range of acoustic measures and also show a small but significant difference in speech patterns. However, these differences do not lead to differences in human or automated scores of English language proficiency. Several measures of bias showed no differences in scores between the two groups.

Most state-of-the-art Deep Learning systems for speaker verification are based on speaker embedding extractors. These architectures are commonly composed of a feature extractor front-end together with a pooling layer to encode variable-length utterances into fixed-length speaker vectors. In this paper we present Double Multi-Head Attention pooling, which extends our previous approach based on Self Multi-Head Attention. An additional self attention layer is added to the pooling layer that summarizes the context vectors produced by Multi-Head Attention into a unique speaker representation. This method enhances the pooling mechanism by giving weights to the information captured for each head and it results in creating more discriminative speaker embeddings. We have evaluated our approach with the VoxCeleb2 dataset. Our results show 6.09% and 5.23% relative improvement in terms of EER compared to Self Attention pooling and Self Multi-Head Attention, respectively. According to the obtained results, Double Multi-Head Attention has shown to be an excellent approach to efficiently select the most relevant features captured by the CNN-based front-ends from the speech signal.

Synthetic creation of drum sounds (e.g., in drum machines) is commonly performed using analog or digital synthesis, allowing a musician to sculpt the desired timbre modifying various parameters. Typically, such parameters control low-level features of the sound and often have no musical meaning or perceptual correspondence. With the rise of Deep Learning, data-driven processing of audio emerges as an alternative to traditional signal processing. This new paradigm allows controlling the synthesis process through learned high-level features or by conditioning a model on musically relevant information. In this paper, we apply a Generative Adversarial Network to the task of audio synthesis of drum sounds. By conditioning the model on perceptual features computed with a publicly available feature-extractor, intuitive control is gained over the generation process. The experiments are carried out on a large collection of kick, snare, and cymbal sounds. We show that, compared to a specific prior work based on a U-Net architecture, our approach considerably improves the quality of the generated drum samples, and that the conditional input indeed shapes the perceptual characteristics of the sounds. Also, we provide audio examples and release the code used in our experiments.

The continuous speech separation (CSS) is a task to separate the speech sources from a long, partially overlapped recording, which involves a varying number of speakers. A straightforward extension of conventional utterance-level speech separation to the CSS task is to segment the long recording with a size-fixed window and process each window separately. Though effective, this extension fails to model the long dependency in speech and thus leads to sub-optimum performance. The recent proposed dual-path modeling could be a remedy to this problem, thanks to its capability in jointly modeling the cross-window dependency and the local-window processing. In this work, we further extend the dual-path modeling framework for CSS task. A transformer-based dual-path system is proposed, which integrates transform layers for global modeling. The proposed models are applied to LibriCSS, a real recorded multi-talk dataset, and consistent WER reduction can be observed in the ASR evaluation for separated speech. Also, a dual-path transformer equipped with convolutional layers is proposed. It significantly reduces the computation amount by 30% with better WER evaluation. Furthermore, the online processing dual-path models are investigated, which shows 10% relative WER reduction compared to the baseline.