Communication technologies like voice over IP operate under constrained real-time conditions, with voice packets being subject to delays and losses from the network. In such cases, the packet loss concealment (PLC) algorithm reconstructs missing fram
es until a new real packet is received. Recently, autoregressive deep neural networks have been shown to surpass the quality of signal processing methods for PLC, specially for long-term predictions beyond 60 ms. In this work, we propose a non-autoregressive adversarial auto-encoder, named PLAAE, to perform real-time PLC in the waveform domain. PLAAE has a causal convolutional structure, and it learns in an auto-encoder fashion to reconstruct signals with gaps, with the help of an adversarial loss. During inference, it is able to predict smooth and coherent continuations of such gaps in a single feed-forward step, as opposed to autoregressive models. Our evaluation highlights the superiority of PLAAE over two classic PLCs and two deep autoregressive models in terms of spectral and intonation reconstruction, perceptual quality, and intelligibility.
Score-based generative models provide state-of-the-art quality for image and audio synthesis. Sampling from these models is performed iteratively, typically employing a discretized series of noise levels and a predefined scheme. In this note, we firs
t overview three common sampling schemes for models trained with denoising score matching. Next, we focus on one of them, consistent annealed sampling, and study its hyper-parameter boundaries. We then highlight a possible formulation of such hyper-parameter that explicitly considers those boundaries and facilitates tuning when using few or a variable number of steps. Finally, we highlight some connections of the formulation with other sampling schemes.
A number of recent advances in neural audio synthesis rely on upsampling layers, which can introduce undesired artifacts. In computer vision, upsampling artifacts have been studied and are known as checkerboard artifacts (due to their characteristic
visual pattern). However, their effect has been overlooked so far in audio processing. Here, we address this gap by studying this problem from the audio signal processing perspective. We first show that the main sources of upsampling artifacts are: (i) the tonal and filtering artifacts introduced by problematic upsampling operators, and (ii) the spectral replicas that emerge while upsampling. We then compare different upsampling layers, showing that nearest neighbor upsamplers can be an alternative to the problematic (but state-of-the-art) transposed and subpixel convolutions which are prone to introduce tonal artifacts.
Applications of deep learning to automatic multitrack mixing are largely unexplored. This is partly due to the limited available data, coupled with the fact that such data is relatively unstructured and variable. To address these challenges, we propo
se a domain-inspired model with a strong inductive bias for the mixing task. We achieve this with the application of pre-trained sub-networks and weight sharing, as well as with a sum/difference stereo loss function. The proposed model can be trained with a limited number of examples, is permutation invariant with respect to the input ordering, and places no limit on the number of input sources. Furthermore, it produces human-readable mixing parameters, allowing users to manually adjust or refine the generated mix. Results from a perceptual evaluation involving audio engineers indicate that our approach generates mixes that outperform baseline approaches. To the best of our knowledge, this work demonstrates the first approach in learning multitrack mixing conventions from real-world data at the waveform level, without knowledge of the underlying mixing parameters.
Automatic speech quality assessment is an important, transversal task whose progress is hampered by the scarcity of human annotations, poor generalization to unseen recording conditions, and a lack of flexibility of existing approaches. In this work,
we tackle these problems with a semi-supervised learning approach, combining available annotations with programmatically generated data, and using 3 different optimization criteria together with 5 complementary auxiliary tasks. Our results show that such a semi-supervised approach can cut the error of existing methods by more than 36%, while providing additional benefits in terms of reusable features or auxiliary outputs. Improvement is further corroborated with an out-of-sample test showing promising generalization capabilities.
Despite the growing interest in unsupervised learning, extracting meaningful knowledge from unlabelled audio remains an open challenge. To take a step in this direction, we recently proposed a problem-agnostic speech encoder (PASE), that combines a c
onvolutional encoder followed by multiple neural networks, called workers, tasked to solve self-supervised problems (i.e., ones that do not require manual annotations as ground truth). PASE was shown to capture relevant speech information, including speaker voice-print and phonemes. This paper proposes PASE+, an improved version of PASE for robust speech recognition in noisy and reverberant environments. To this end, we employ an online speech distortion module, that contaminates the input signals with a variety of random disturbances. We then propose a revised encoder that better learns short- and long-term speech dynamics with an efficient combination of recurrent and convolutional networks. Finally, we refine the set of workers used in self-supervision to encourage better cooperation. Results on TIMIT, DIRHA and CHiME-5 show that PASE+ significantly outperforms both the previous version of PASE as well as common acoustic features. Interestingly, PASE+ learns transferable representations suitable for highly mismatched acoustic conditions.
In many applications of multi-microphone multi-device processing, the synchronization among different input channels can be affected by the lack of a common clock and isolated drops of samples. In this work, we address the issue of sample drop detect
ion in the context of a conversational speech scenario, recorded by a set of microphones distributed in space. The goal is to design a neural-based model that given a short window in the time domain, detects whether one or more devices have been subjected to a sample drop event. The candidate time windows are selected from a set of large time intervals, possibly including a sample drop, and by using a preprocessing step. The latter is based on the application of normalized cross-correlation between signals acquired by different devices. The architecture of the neural network relies on a CNN-LSTM encoder, followed by multi-head attention. The experiments are conducted using both artificial and real data. Our proposed approach obtained F1 score of 88% on an evaluation set extracted from the CHiME-5 corpus. A comparable performance was found in a larger set of experiments conducted on a set of multi-channel artificial scenes.
End-to-end models for raw audio generation are a challenge, specially if they have to work with non-parallel data, which is a desirable setup in many situations. Voice conversion, in which a model has to impersonate a speaker in a recording, is one o
f those situations. In this paper, we propose Blow, a single-scale normalizing flow using hypernetwork conditioning to perform many-to-many voice conversion between raw audio. Blow is trained end-to-end, with non-parallel data, on a frame-by-frame basis using a single speaker identifier. We show that Blow compares favorably to existing flow-based architectures and other competitive baselines, obtaining equal or better performance in both objective and subjective evaluations. We further assess the impact of its main components with an ablation study, and quantify a number of properties such as the necessary amount of training data or the preference for source or target speakers.
Text-to-speech (TTS) acoustic models map linguistic features into an acoustic representation out of which an audible waveform is generated. The latest and most natural TTS systems build a direct mapping between linguistic and waveform domains, like S
ampleRNN. This way, possible signal naturalness losses are avoided as intermediate acoustic representations are discarded. Another important dimension of study apart from naturalness is their adaptability to generate voice from new speakers that were unseen during training. In this paper we first propose the use of problem-agnostic speech embeddings in a multi-speaker acoustic model for TTS based on SampleRNN. This way we feed the acoustic model with speaker acoustically dependent representations that enrich the waveform generation more than discrete embeddings unrelated to these factors. Our first results suggest that the proposed embeddings lead to better quality voices than those obtained with discrete embeddings. Furthermore, as we can use any speech segment as an encoded representation during inference, the model is capable to generalize to new speaker identities without retraining the network. We finally show that, with a small increase of speech duration in the embedding extractor, we dramatically reduce the spectral distortion to close the gap towards the target identities.
The speech enhancement task usually consists of removing additive noise or reverberation that partially mask spoken utterances, affecting their intelligibility. However, little attention is drawn to other, perhaps more aggressive signal distortions l
ike clipping, chunk elimination, or frequency-band removal. Such distortions can have a large impact not only on intelligibility, but also on naturalness or even speaker identity, and require of careful signal reconstruction. In this work, we give full consideration to this generalized speech enhancement task, and show it can be tackled with a time-domain generative adversarial network (GAN). In particular, we extend a previous GAN-based speech enhancement system to deal with mixtures of four types of aggressive distortions. Firstly, we propose the addition of an adversarial acoustic regression loss that promotes a richer feature extraction at the discriminator. Secondly, we also make use of a two-step adversarial training schedule, acting as a warm up-and-fine-tune sequence. Both objective and subjective evaluations show that these two additions bring improved speech reconstructions that better match the original speaker identity and naturalness.
