arXiv subjects

Jean-Marc Valin

Publications and source records attributed to Jean-Marc Valin.

At least 37 records · Page 2Linked to original sources

arXiv2021

Personalized PercepNet: Real-time, Low-complexity Target Voice Separation and Enhancement

The presence of multiple talkers in the surrounding environment poses a difficult challenge for real-time speech communication systems considering the constraints on network size and complexity. In this paper, we present Personalized PercepNet, a real-time speech enhancement model that separates a target speaker from a noisy multi-talker mixture without compromising on complexity of the recently proposed PercepNet. To enable speaker-dependent speech enhancement, we first show how we can train a perceptually motivated speaker embedder network to produce a representative embedding vector for the given speaker. Personalized PercepNet uses the target speaker embedding as additional information to pick out and enhance only the target speaker while suppressing all other competing sounds. Our experiments show that the proposed model significantly outperforms PercepNet and other baselines, both in terms of objective speech enhancement metrics and human opinion scores.

eess.AS↗

arXiv2021

Semi-Supervised Singing Voice Separation with Noisy Self-Training

Recent progress in singing voice separation has primarily focused on supervised deep learning methods. However, the scarcity of ground-truth data with clean musical sources has been a problem for long. Given a limited set of labeled data, we present a method to leverage a large volume of unlabeled data to improve the model's performance. Following the noisy self-training framework, we first train a teacher network on the small labeled dataset and infer pseudo-labels from the large corpus of unlabeled mixtures. Then, a larger student network is trained on combined ground-truth and self-labeled datasets. Empirical results show that the proposed self-training scheme, along with data augmentation methods, effectively leverage the large unlabeled corpus and obtain superior performance compared to supervised methods.

eess.AS↗

arXiv2021

Enhancing into the codec: Noise Robust Speech Coding with Vector-Quantized Autoencoders

Audio codecs based on discretized neural autoencoders have recently been developed and shown to provide significantly higher compression levels for comparable quality speech output. However, these models are tightly coupled with speech content, and produce unintended outputs in noisy conditions. Based on VQ-VAE autoencoders with WaveRNN decoders, we develop compressor-enhancer encoders and accompanying decoders, and show that they operate well in noisy conditions. We also observe that a compressor-enhancer model performs better on clean speech inputs than a compressor model trained only on clean speech.

eess.AS↗

arXiv2021

Low-Complexity, Real-Time Joint Neural Echo Control and Speech Enhancement Based On PercepNet

Speech enhancement algorithms based on deep learning have greatly surpassed their traditional counterparts and are now being considered for the task of removing acoustic echo from hands-free communication systems. This is a challenging problem due to both real-world constraints like loudspeaker non-linearities, and to limited compute capabilities in some communication systems. In this work, we propose a system combining a traditional acoustic echo canceller, and a low-complexity joint residual echo and noise suppressor based on a hybrid signal processing/deep neural network (DSP/DNN) approach. We show that the proposed system outperforms both traditional and other neural approaches, while requiring only 5.5% CPU for real-time operation. We further show that the system can scale to even lower complexity levels.

eess.AS↗

arXiv2020

A Perceptually-Motivated Approach for Low-Complexity, Real-Time Enhancement of Fullband Speech

Over the past few years, speech enhancement methods based on deep learning have greatly surpassed traditional methods based on spectral subtraction and spectral estimation. Many of these new techniques operate directly in the the short-time Fourier transform (STFT) domain, resulting in a high computational complexity. In this work, we propose PercepNet, an efficient approach that relies on human perception of speech by focusing on the spectral envelope and on the periodicity of the speech. We demonstrate high-quality, real-time enhancement of fullband (48 kHz) speech with less than 5% of a CPU core.

eess.AS↗

arXiv2020

PoCoNet: Better Speech Enhancement with Frequency-Positional Embeddings, Semi-Supervised Conversational Data, and Biased Loss

Neural network applications generally benefit from larger-sized models, but for current speech enhancement models, larger scale networks often suffer from decreased robustness to the variety of real-world use cases beyond what is encountered in training data. We introduce several innovations that lead to better large neural networks for speech enhancement. The novel PoCoNet architecture is a convolutional neural network that, with the use of frequency-positional embeddings, is able to more efficiently build frequency-dependent features in the early layers. A semi-supervised method helps increase the amount of conversational training data by pre-enhancing noisy datasets, improving performance on real recordings. A new loss function biased towards preserving speech quality helps the optimization better match human perceptual opinions on speech quality. Ablation experiments and objective and human opinion metrics show the benefits of the proposed improvements.

eess.AS↗

arXiv2020

Improving Opus Low Bit Rate Quality with Neural Speech Synthesis

The voice mode of the Opus audio coder can compress wideband speech at bit rates ranging from 6 kb/s to 40 kb/s. However, Opus is at its core a waveform matching coder, and as the rate drops below 10 kb/s, quality degrades quickly. As the rate reduces even further, parametric coders tend to perform better than waveform coders. In this paper we propose a backward-compatible way of improving low bit rate Opus quality by re-synthesizing speech from the decoded parameters. We compare two different neural generative models, WaveNet and LPCNet. WaveNet is a powerful, high-complexity, and high-latency architecture that is not feasible for a practical system, yet provides a best known achievable quality with generative models. LPCNet is a low-complexity, low-latency RNN-based generative model, and practically implementable on mobile phones. We apply these systems with parameters from Opus coded at 6 kb/s as conditioning features for the generative models. A listening test shows that for the same 6 kb/s Opus bit stream, synthesized speech using LPCNet clearly outperforms the output of the standard Opus decoder. This opens up ways to improve the decoding quality of existing speech and audio waveform coders without breaking compatibility.

eess.AS↗

arXiv2019

A Real-Time Wideband Neural Vocoder at 1.6 kb/s Using LPCNet

Neural speech synthesis algorithms are a promising new approach for coding speech at very low bitrate. They have so far demonstrated quality that far exceeds traditional vocoders, at the cost of very high complexity. In this work, we present a low-bitrate neural vocoder based on the LPCNet model. The use of linear prediction and sparse recurrent networks makes it possible to achieve real-time operation on general-purpose hardware. We demonstrate that LPCNet operating at 1.6 kb/s achieves significantly higher quality than MELP and that uncompressed LPCNet can exceed the quality of a waveform codec operating at low bitrate. This opens the way for new codec designs based on neural synthesis models.

eess.AS↗

arXiv2019

LPCNet: Improving Neural Speech Synthesis Through Linear Prediction

Neural speech synthesis models have recently demonstrated the ability to synthesize high quality speech for text-to-speech and compression applications. These new models often require powerful GPUs to achieve real-time operation, so being able to reduce their complexity would open the way for many new applications. We propose LPCNet, a WaveRNN variant that combines linear prediction with recurrent neural networks to significantly improve the efficiency of speech synthesis. We demonstrate that LPCNet can achieve significantly higher quality than WaveRNN for the same network size and that high quality LPCNet speech synthesis is achievable with a complexity under 3 GFLOPS. This makes it easier to deploy neural synthesis applications on lower-power devices, such as embedded systems and mobile phones.

eess.AS↗

arXiv2018

A Hybrid DSP/Deep Learning Approach to Real-Time Full-Band Speech Enhancement

Despite noise suppression being a mature area in signal processing, it remains highly dependent on fine tuning of estimator algorithms and parameters. In this paper, we demonstrate a hybrid DSP/deep learning approach to noise suppression. A deep neural network with four hidden layers is used to estimate ideal critical band gains, while a more traditional pitch filter attenuates noise between pitch harmonics. The approach achieves significantly higher quality than a traditional minimum mean squared error spectral estimator, while keeping the complexity low enough for real-time operation at 48 kHz on a low-power processor.

cs.SD↗

arXiv2017

The AV1 Constrained Directional Enhancement Filter (CDEF)

This paper presents the constrained directional enhancement filter designed for the AV1 royalty-free video codec. The in-loop filter is based on a non-linear low-pass filter and is designed for vectorization efficiency. It takes into account the direction of edges and patterns being filtered. The filter works by identifying the direction of each block and then adaptively filtering with a high degree of control over the filter strength along the direction and across it. The proposed enhancement filter is shown to improve the quality of the Alliance for Open Media (AOM) AV1 and Thor video codecs in particular in low complexity configurations.

cs.MM↗

arXiv2016

Perceptually-Driven Video Coding with the Daala Video Codec

The Daala project is a royalty-free video codec that attempts to compete with the best patent-encumbered codecs. Part of our strategy is to replace core tools of traditional video codecs with alternative approaches, many of them designed to take perceptual aspects into account, rather than optimizing for simple metrics like PSNR. This paper documents some of our experiences with these tools, which ones worked and which did not. We evaluate which tools are easy to integrate into a more traditional codec design, and show results in the context of the codec being developed by the Alliance for Open Media.

cs.MM↗

arXiv2016

Daala: Building A Next-Generation Video Codec From Unconventional Technology

Daala is a new royalty-free video codec that attempts to compete with state-of-the-art royalty-bearing codecs. To do so, it must achieve good compression while avoiding all of their patented techniques. We use technology that is as different as possible from traditional approaches to achieve this. This paper describes the technology behind Daala and discusses where it fits in the newly created AV1 codec from the Alliance for Open Media. We show that Daala is approaching the performance level of more mature, state-of-the art video codecs and can contribute to improving AV1.

cs.MM↗

arXiv2016

Daala: A Perceptually-Driven Still Picture Codec

Daala is a new royalty-free video codec based on perceptually-driven coding techniques. We explore using its keyframe format for still picture coding and show how it has improved over the past year. We believe the technology used in Daala could be the basis of an excellent, royalty-free image format.

cs.MM↗

arXiv2016

Daala: A Perceptually-Driven Next Generation Video Codec

The Daala project is a royalty-free video codec that attempts to compete with the best patent-encumbered codecs. Part of our strategy is to replace core tools of traditional video codecs with alternative approaches, many of them designed to take perceptual aspects into account, rather than optimizing for simple metrics like PSNR. This paper documents some of our experiences with these tools, which ones worked and which did not, and what we've learned from them. The result is a codec which compares favorably with HEVC on still images, and is on a path to do so for video as well.

cs.MM↗

arXiv2016

Channel Decorrelation For Stereo Acoustic Echo Cancellation In High-Quality Audio Communication

In this paper, we address an important problem in high-quality audio communication systems. Acoustic echo cancellation with stereo signals is generally an under-determined problem because of the generally important correlation that exists between the left and right channels. In this paper, we present a novel method of significantly reducing that correlation without affecting the audio quality. This method is perceptually motivated and combines a shaped comb-allpass (SCAL) filter with the injection of psychoacoustically masked noise. We show that the proposed method performs significantly better than other known methods for channel decorrelation.

cs.SD↗

arXiv2016

Microphone array post-filter for separation of simultaneous non-stationary sources

Microphone array post-filters have demonstrated their ability to greatly reduce noise at the output of a beamformer. However, current techniques only consider a single source of interest, most of the time assuming stationary background noise. We propose a microphone array post-filter that enhances the signals produced by the separation of simultaneous sources using common source separation algorithms. Our method is based on a loudness-domain optimal spectral estimator and on the assumption that the noise can be described as the sum of a stationary component and of a transient component that is due to leakage between the channels of the initial source separation algorithm. The system is evaluated in the context of mobile robotics and is shown to produce better results than current post-filtering techniques, greatly reducing interference while causing little distortion to the signal of interest, even at very low SNR.

cs.SD↗

arXiv2016

Predicting Chroma from Luma with Frequency Domain Intra Prediction

This paper describes a technique for performing intra prediction of the chroma planes based on the reconstructed luma plane in the frequency domain. This prediction exploits the fact that while RGB to YUV color conversion has the property that it decorrelates the color planes globally across an image, there is still some correlation locally at the block level. Previous proposals compute a linear model of the spatial relationship between the luma plane (Y) and the two chroma planes (U and V). In codecs that use lapped transforms this is not possible since transform support extends across the block boundaries and thus neighboring blocks are unavailable during intra-prediction. We design a frequency domain intra predictor for chroma that exploits the same local correlation with lower complexity than the spatial predictor and which works with lapped transforms. We then describe a low-complexity algorithm that directly uses luma coefficients as a chroma predictor based on gain-shape quantization and band partitioning. An experiment is performed that compares these two techniques inside the experimental Daala video codec and shows the lower complexity algorithm to be a better chroma predictor.

cs.MM↗