This paper proposes a generative pretraining foundation model for high-quality speech restoration tasks. By directly operating on complex-valued short-time Fourier transform coefficients, our model does not rely on any vocoders for time-domain signal reconstruction. As a result, our model simplifies the synthesis process and removes the quality upper-bound introduced by any mel-spectrogram vocoder compared to prior work SpeechFlow. The proposed method is evaluated on multiple speech restoration tasks, including speech denoising, bandwidth extension, codec artifact removal, and target speaker extraction. In all scenarios, finetuning our pretrained model results in superior performance over strong baselines. Notably, in the target speaker extraction task, our model outperforms existing systems, including those leveraging SSL-pretrained encoders like WavLM. The code and the pretrained checkpoints are publicly available in the NVIDIA NeMo framework.
Vibravox is a dataset compliant with the General Data Protection Regulation (GDPR) containing audio recordings using five different body-conduction audio sensors: two in-ear microphones, two bone conduction vibration pickups, and a laryngophone. The dataset also includes audio data from an airborne microphone used as a reference. The Vibravox corpus contains 45 hours per sensor of speech samples and physiological sounds recorded by 188 participants under different acoustic conditions imposed by a high order ambisonics 3D spatializer. Annotations about the recording conditions and linguistic transcriptions are also included in the corpus. We conducted a series of experiments on various speech-related tasks, including speech recognition, speech enhancement, and speaker verification. These experiments were carried out using state-of-the-art models to evaluate and compare their performances on signals captured by the different audio sensors offered by the Vibravox dataset, with the aim of gaining a better grasp of their individual characteristics.
The majority of existing speech bandwidth extension (BWE) methods operate under the constraint of fixed source and target sampling rates, which limits their flexibility in practical applications. In this paper, we propose a multi-stage speech BWE model named MS-BWE, which can handle a set of source and target sampling rate pairs and achieve flexible extensions of frequency bandwidth. The proposed MS-BWE model comprises a cascade of BWE blocks, with each block featuring a dual-stream architecture to realize amplitude and phase extension, progressively painting the speech frequency bands stage by stage. The teacher-forcing strategy is employed to mitigate the discrepancy between training and inference. Experimental results demonstrate that our proposed MS-BWE is comparable to state-of-the-art speech BWE methods in speech quality. Regarding generation efficiency, the one-stage generation of MS-BWE can achieve over one thousand times real-time on GPU and about sixty times on CPU.
This paper presents a novel approach to audio restoration, focusing on the enhancement of low-quality music recordings, and in particular historical ones. Building upon a previous algorithm called BABE, or Blind Audio Bandwidth Extension, we introduce BABE-2, which presents a series of improvements. This research broadens the concept of bandwidth extension to \emph{generative equalization}, a novel task that, to the best of our knowledge, has not been explicitly addressed in previous studies. BABE-2 is built around an optimization algorithm utilizing priors from diffusion models, which are trained or fine-tuned using a curated set of high-quality music tracks. The algorithm simultaneously performs two critical tasks: estimation of the filter degradation magnitude response and hallucination of the restored audio. The proposed method is objectively evaluated on historical piano recordings, showing an enhancement over the prior version. The method yields similarly impressive results in rejuvenating the works of renowned vocalists Enrico Caruso and Nellie Melba. This research represents an advancement in the practical restoration of historical music.
Speech bandwidth extension (BWE) refers to widening the frequency bandwidth range of speech signals, enhancing the speech quality towards brighter and fuller. This paper proposes a generative adversarial network (GAN) based BWE model with parallel prediction of Amplitude and Phase spectra, named AP-BWE, which achieves both high-quality and efficient wideband speech waveform generation. The proposed AP-BWE generator is entirely based on convolutional neural networks (CNNs). It features a dual-stream architecture with mutual interaction, where the amplitude stream and the phase stream communicate with each other and respectively extend the high-frequency components from the input narrowband amplitude and phase spectra. To improve the naturalness of the extended speech signals, we employ a multi-period discriminator at the waveform level and design a pair of multi-resolution amplitude and phase discriminators at the spectral level, respectively. Experimental results demonstrate that our proposed AP-BWE achieves state-of-the-art performance in terms of speech quality for BWE tasks targeting sampling rates of both 16 kHz and 48 kHz. In terms of generation efficiency, due to the all-convolutional architecture and all-frame-level operations, the proposed AP-BWE can generate 48 kHz waveform samples 292.3 times faster than real-time on a single RTX 4090 GPU and 18.1 times faster than real-time on a single CPU. Notably, to our knowledge, AP-BWE is the first to achieve the direct extension of the high-frequency phase spectrum, which is beneficial for improving the effectiveness of existing BWE methods.
This paper presents a configurable version of Extreme Bandwidth Extension Network (EBEN), a Generative Adversarial Network (GAN) designed to improve audio captured with body-conduction microphones. We show that although these microphones significantly reduce environmental noise, this insensitivity to ambient noise happens at the expense of the bandwidth of the speech signal acquired by the wearer of the devices. The obtained captured signals therefore require the use of signal enhancement techniques to recover the full-bandwidth speech. EBEN leverages a configurable multiband decomposition of the raw captured signal. This decomposition allows the data time domain dimensions to be reduced and the full band signal to be better controlled. The multiband representation of the captured signal is processed through a U-Net-like model, which combines feature and adversarial losses to generate an enhanced speech signal. We also benefit from this original representation in the proposed configurable discriminators architecture. The configurable EBEN approach can achieve state-of-the-art enhancement results on synthetic data with a lightweight generator that allows real-time processing.
We present AERO, a audio super-resolution model that processes speech and music signals in the spectral domain. AERO is based on an encoder-decoder architecture with U-Net like skip connections. We optimize the model using both time and frequency domain loss functions. Specifically, we consider a set of reconstruction losses together with perceptual ones in the form of adversarial and feature discriminator loss functions. To better handle phase information the proposed method operates over the complex-valued spectrogram using two separate channels. Unlike prior work which mainly considers low and high frequency concatenation for audio super-resolution, the proposed method directly predicts the full frequency range. We demonstrate high performance across a wide range of sample rates considering both speech and music. AERO outperforms the evaluated baselines considering Log-Spectral Distance, ViSQOL, and the subjective MUSHRA test. Audio samples and code are available at https://pages.cs.huji.ac.il/adiyoss-lab/aero
In this paper, we present Extreme Bandwidth Extension Network (EBEN), a Generative Adversarial network (GAN) that enhances audio measured with body-conduction microphones. This type of capture equipment suppresses ambient noise at the expense of speech bandwidth, thereby requiring signal enhancement techniques to recover the wideband speech signal. EBEN leverages a multiband decomposition of the raw captured speech to decrease the data time-domain dimensions, and give better control over the full-band signal. This multiband representation is fed to a U-Net-like model, which adopts a combination of feature and adversarial losses to recover an enhanced audio signal. We also benefit from this original representation in the proposed discriminator architecture. Our approach can achieve state-of-the-art results with a lightweight generator and real-time compatible operation.
This paper presents CQT-Diff, a data-driven generative audio model that can, once trained, be used for solving various different audio inverse problems in a problem-agnostic setting. CQT-Diff is a neural diffusion model with an architecture that is carefully constructed to exploit pitch-equivariant symmetries in music. This is achieved by preconditioning the model with an invertible Constant-Q Transform (CQT), whose logarithmically-spaced frequency axis represents pitch equivariance as translation equivariance. The proposed method is evaluated with objective and subjective metrics in three different and varied tasks: audio bandwidth extension, inpainting, and declipping. The results show that CQT-Diff outperforms the compared baselines and ablations in audio bandwidth extension and, without retraining, delivers competitive performance against modern baselines in audio inpainting and declipping. This work represents the first diffusion-based general framework for solving inverse problems in audio processing.
Audio bandwidth extension aims to expand the spectrum of narrow-band audio signals. Although this topic has been broadly studied during recent years, the particular problem of extending the bandwidth of historical music recordings remains an open challenge. This paper proposes BEHM-GAN, a model based on generative adversarial networks, as a practical solution to this problem. The proposed method works with the complex spectrogram representation of audio and, thanks to a dedicated regularization strategy, can effectively extend the bandwidth of out-of-distribution real historical recordings. The BEHM-GAN is designed to be applied as a second step after denoising the recording to suppress any additive disturbances, such as clicks and background noise. We train and evaluate the method using solo piano classical music. The proposed method outperforms the compared baselines in both objective and subjective experiments. The results of a formal blind listening test show that BEHM-GAN significantly increases the perceptual sound quality in early-20th-century gramophone recordings. For several items, there is a substantial improvement in the mean opinion score after enhancing historical recordings with the proposed bandwidth-extension algorithm. This study represents a relevant step toward data-driven music restoration in real-world scenarios.
Speech super-resolution (SR) is a task to increase speech sampling rate by generating high-frequency components. Existing speech SR methods are trained in constrained experimental settings, such as a fixed upsampling ratio. These strong constraints can potentially lead to poor generalization ability in mismatched real-world cases. In this paper, we propose a neural vocoder based speech super-resolution method (NVSR) that can handle a variety of input resolution and upsampling ratios. NVSR consists of a mel-bandwidth extension module, a neural vocoder module, and a post-processing module. Our proposed system achieves state-of-the-art results on the VCTK multi-speaker benchmark. On 44.1 kHz target resolution, NVSR outperforms WSRGlow and Nu-wave by 8% and 37% respectively on log spectral distance and achieves a significantly better perceptual quality. We also demonstrate that prior knowledge in the pre-trained vocoder is crucial for speech SR by performing mel-bandwidth extension with a simple replication-padding method. Samples can be found in https://haoheliu.github.io/nvsr.
Generative adversarial networks have recently demonstrated outstanding performance in neural vocoding outperforming best autoregressive and flow-based models. In this paper, we show that this success can be extended to other tasks of conditional audio generation. In particular, building upon HiFi vocoders, we propose a novel HiFi++ general framework for bandwidth extension and speech enhancement. We show that with the improved generator architecture, HiFi++ performs better or comparably with the state-of-the-art in these tasks while spending significantly less computational resources. The effectiveness of our approach is validated through a series of extensive experiments.
We introduce a block-online variant of the temporal feature-wise linear modulation (TFiLM) model to achieve bandwidth extension. The proposed architecture simplifies the UNet backbone of the TFiLM to reduce inference time and employs an efficient transformer at the bottleneck to alleviate performance degradation. We also utilize self-supervised pretraining and data augmentation to enhance the quality of bandwidth extended signals and reduce the sensitivity with respect to downsampling methods. Experiment results on the VCTK dataset show that the proposed method outperforms several recent baselines in both intrusive and non-intrusive metrics. Pretraining and filter augmentation also help stabilize and enhance the overall performance.
In this paper, we address a sub-topic of the broad domain of audio enhancement, namely musical audio bandwidth extension. We formulate the bandwidth extension problem using deep neural networks, where a band-limited signal is provided as input to the network, with the goal of reconstructing a full-bandwidth output. Our main contribution centers on the impact of the choice of low pass filter when training and subsequently testing the network. For two different state of the art deep architectures, ResNet and U-Net, we demonstrate that when the training and testing filters are matched, improvements in signal-to-noise ratio (SNR) of up to 7dB can be obtained. However, when these filters differ, the improvement falls considerably and under some training conditions results in a lower SNR than the band-limited input. To circumvent this apparent overfitting to filter shape, we propose a data augmentation strategy which utilizes multiple low pass filters during training and leads to improved generalization to unseen filtering conditions at test time.