Language models have been successfully used to model natural signals, such as images, speech, and music. A key component of these models is a high quality neural compression model that can compress high-dimensional natural signals into lower dimensional discrete tokens. To that end, we introduce a high-fidelity universal neural audio compression algorithm that achieves ~90x compression of 44.1 KHz audio into tokens at just 8kbps bandwidth. We achieve this by combining advances in high-fidelity audio generation with better vector quantization techniques from the image domain, along with improved adversarial and reconstruction losses. We compress all domains (speech, environment, music, etc.) with a single universal model, making it widely applicable to generative modeling of all audio. We compare with competing audio compression algorithms, and find our method outperforms them significantly. We provide thorough ablations for every design choice, as well as open-source code and trained model weights. We hope our work can lay the foundation for the next generation of high-fidelity audio modeling.

We presentSoundStream, a novel neural audio codec that can efficiently compress speech, music and general audio at bitrates normally targeted by speech-tailored codecs.SoundStreamrelies on a model architecture composed by a fully convolutional encoder/decoder network and a residual vector quantizer, which are trained jointly end-to-end. Training leverages recent advances in text-to-speech and speech enhancement, which combine adversarial and reconstruction losses to allow the generation of high-quality audio content from quantized embeddings. By training with structured dropout applied to quantizer layers, a single model can operate across variable bitrates from 3 kbps to 18 kbps, with a negligible quality loss when compared with models trained at fixed bitrates. In addition, the model is amenable to a low latency implementation, which supports streamable inference and runs in real time on a smartphone CPU. In subjective evaluations using audio at 24 kHz sampling rate,SoundStreamat 3 kbps outperforms Opus at 12 kbps and approaches EVS at 9.6 kbps. Moreover, we are able to perform joint compression and enhancement either at the encoder or at the decoder side with no additional latency, which we demonstrate through background noise suppression for speech.

The recent emergence of machine-learning based generative models for speech suggests a significant reduction in bit rate for speech codecs is possible. However, the performance of generative models deteriorates significantly with the distortions present in real-world input signals. We argue that this deterioration is due to the sensitivity of the maximum likelihood criterion to outliers and the ineffectiveness of modeling a sum of independent signals with a single autoregressive model. We introduce predictive-variance regularization to reduce the sensitivity to outliers, resulting in a significant increase in performance. We show that noise reduction to remove unwanted signals can significantly increase performance. We provide extensive subjective performance evaluations that show that our system based on generative modeling provides state-of-the-art coding performance at 3 kb/s for real-world speech signals at reasonable computational complexity.

In order to efficiently transmit and store speech signals, speech codecs create a minimally redundant representation of the input signal which is then decoded at the receiver with the best possible perceptual quality. In this work we demonstrate that a neural network architecture based on VQ-VAE with a WaveNet decoder can be used to perform very low bit-rate speech coding with high reconstruction quality. A prosody-transparent and speaker-independent model trained on the LibriSpeech corpus coding audio at 1.6 kbps exhibits perceptual quality which is around halfway between the MELP codec at 2.4 kbps and AMR-WB codec at 23.05 kbps. In addition, when training on high-quality recorded speech with the test speaker included in the training set, a model coding speech at 1.6 kbps produces output of similar perceptual quality to that generated by AMR-WB at 23.05 kbps.

Traditional parametric coding of speech facilitates low rate but provides poor reconstruction quality because of the inadequacy of the model used. We describe how a WaveNet generative speech model can be used to generate high quality speech from the bit stream of a standard parametric coder operating at 2.4 kb/s. We compare this parametric coder with a waveform coder based on the same generative model and show that approximating the signal waveform incurs a large rate penalty. Our experiments confirm the high performance of the WaveNet based coder and show that the speech produced by the system is able to additionally perform implicit bandwidth extension and does not significantly impair recognition of the original speaker for the human listener, even when that speaker has not been used during the training of the generative model.

We address the problem of acoustic beamforming with a small number of microphones given only limited knowledge of the spatial scenario. We first identify a set of plausible target (desired source) scenarios and a set of interferer scenarios that have desirable suppression characteristics. We then design soft masks (postprocessors) for all target-interferer scenario pairs and select the composite scenario that maximizes the output variance over target scenarios and minimizes it over interferer scenarios. This corresponds to an approximate concatenation of soft masks. The result is a nonlinear beamformer with an adjustable beamwidth and an adjustable region where point interferers are strongly suppressed, even for two microphones. For the individual masks we use a new postprocessor formulation that is robust to scenario mismatch. The resulting system provides excellent performance with few microphones even when the acoustic scenario is ill-defined.

This paper presents a practically efficient implementation for non-linear acoustic echo cancellation (NAEC). The echo path is modeled by a novel hybrid Taylor-Volterra pre-processor followed by a linear FIR filter. A cascaded block RLS and unconstrained FLMS adaptive algorithm is developed to jointly identify the pre-processor and the FIR filter. This implementation is validated via simulations.

Existing post-filtering techniques for microphone array speech enhancement have two common deficiencies. First, they assume that the noise is either white or diffuse and cannot deal with point inter-ferers. Second, they estimate the post-filter coefficients using only two microphones at a time and then perform averaging over all microphone pairs, yielding a suboptimal solution at best. In this paper, we present a novel post-filtering algorithm that alleviates the first limitation by using a more generalized signal model including not only white and diffuse but also point interferers, and overcomes the second deficiency by offering a globally optimized least-squares solution over all microphones. It is shown by simulations that the proposed method outperforms the existing algorithms in many different acoustic scenarios.