Christian J. Steinmetz

Joseph Turian, Jordie Shier, Humair Raj Khan et al. · cmu

What audio embedding approach generalizes best to a wide range of downstream tasks across a variety of everyday domains without fine-tuning? The aim of the HEAR benchmark is to develop a general-purpose audio representation that provides a strong basis for learning in a wide variety of tasks and scenarios. HEAR evaluates audio representations using a benchmark suite across a variety of domains, including speech, environmental sound, and music. HEAR was launched as a NeurIPS 2021 shared challenge. In the spirit of shared exchange, each participant submitted an audio embedding model following a common API that is general-purpose, open-source, and freely available to use. Twenty-nine models by thirteen external teams were evaluated on nineteen diverse downstream tasks derived from sixteen datasets. Open evaluation code, submitted models and datasets are key contributions, enabling comprehensive and reproducible evaluation, as well as previously impossible longitudinal studies. It still remains an open question whether one single general-purpose audio representation can perform as holistically as the human ear.

Shubhr Singh, Christian J. Steinmetz, Emmanouil Benetos et al.

Deep learning models such as CNNs and Transformers have achieved impressive performance for end-to-end audio tagging. Recent works have shown that despite stacking multiple layers, the receptive field of CNNs remains severely limited. Transformers on the other hand are able to map global context through self-attention, but treat the spectrogram as a sequence of patches which is not flexible enough to capture irregular audio objects. In this work, we treat the spectrogram in a more flexible way by considering it as graph structure and process it with a novel graph neural architecture called ATGNN. ATGNN not only combines the capability of CNNs with the global information sharing ability of Graph Neural Networks, but also maps semantic relationships between learnable class embeddings and corresponding spectrogram regions. We evaluate ATGNN on two audio tagging tasks, where it achieves 0.585 mAP on the FSD50K dataset and 0.335 mAP on the AudioSet-balanced dataset, achieving comparable results to Transformer based models with significantly lower number of learnable parameters.

Marco Comunità, Christian J. Steinmetz, Huy Phan et al.

Deep learning approaches for black-box modelling of audio effects have shown promise, however, the majority of existing work focuses on nonlinear effects with behaviour on relatively short time-scales, such as guitar amplifiers and distortion. While recurrent and convolutional architectures can theoretically be extended to capture behaviour at longer time scales, we show that simply scaling the width, depth, or dilation factor of existing architectures does not result in satisfactory performance when modelling audio effects such as fuzz and dynamic range compression. To address this, we propose the integration of time-varying feature-wise linear modulation into existing temporal convolutional backbones, an approach that enables learnable adaptation of the intermediate activations. We demonstrate that our approach more accurately captures long-range dependencies for a range of fuzz and compressor implementations across both time and frequency domain metrics. We provide sound examples, source code, and pretrained models to faciliate reproducibility.

Christopher Mitcheltree, Christian J. Steinmetz, Marco Comunità et al.

Low frequency oscillator (LFO) driven audio effects such as phaser, flanger, and chorus, modify an input signal using time-varying filters and delays, resulting in characteristic sweeping or widening effects. It has been shown that these effects can be modeled using neural networks when conditioned with the ground truth LFO signal. However, in most cases, the LFO signal is not accessible and measurement from the audio signal is nontrivial, hindering the modeling process. To address this, we propose a framework capable of extracting arbitrary LFO signals from processed audio across multiple digital audio effects, parameter settings, and instrument configurations. Since our system imposes no restrictions on the LFO signal shape, we demonstrate its ability to extract quasiperiodic, combined, and distorted modulation signals that are relevant to effect modeling. Furthermore, we show how coupling the extraction model with a simple processing network enables training of end-to-end black-box models of unseen analog or digital LFO-driven audio effects using only dry and wet audio pairs, overcoming the need to access the audio effect or internal LFO signal. We make our code available and provide the trained audio effect models in a real-time VST plugin.

Hanzhi Yin, Gang Cheng, Christian J. Steinmetz et al.

We describe a novel approach for developing realistic digital models of dynamic range compressors for digital audio production by analyzing their analog prototypes. While realistic digital dynamic compressors are potentially useful for many applications, the design process is challenging because the compressors operate nonlinearly over long time scales. Our approach is based on the structured state space sequence model (S4), as implementing the state-space model (SSM) has proven to be efficient at learning long-range dependencies and is promising for modeling dynamic range compressors. We present in this paper a deep learning model with S4 layers to model the Teletronix LA-2A analog dynamic range compressor. The model is causal, executes efficiently in real time, and achieves roughly the same quality as previous deep-learning models but with fewer parameters.

Christian J. Steinmetz, Joshua D. Reiss

Applications of deep learning for audio effects often focus on modeling analog effects or learning to control effects to emulate a trained audio engineer. However, deep learning approaches also have the potential to expand creativity through neural audio effects that enable new sound transformations. While recent work demonstrated that neural networks with random weights produce compelling audio effects, control of these effects is limited and unintuitive. To address this, we introduce a method for the steerable discovery of neural audio effects. This method enables the design of effects using example recordings provided by the user. We demonstrate how this method produces an effect similar to the target effect, along with interesting inaccuracies, while also providing perceptually relevant controls.

Joseph T. Colonel, Christian J. Steinmetz, Marcus Michelen et al.

Designing infinite impulse response filters to match an arbitrary magnitude response requires specialized techniques. Methods like modified Yule-Walker are relatively efficient, but may not be sufficiently accurate in matching high order responses. On the other hand, iterative optimization techniques often enable superior performance, but come at the cost of longer run-times and are sensitive to initial conditions, requiring manual tuning. In this work, we address some of these limitations by learning a direct mapping from the target magnitude response to the filter coefficient space with a neural network trained on millions of random filters. We demonstrate our approach enables both fast and accurate estimation of filter coefficients given a desired response. We investigate training with different families of random filters, and find training with a variety of filter families enables better generalization when estimating real-world filters, using head-related transfer functions and guitar cabinets as case studies. We compare our method against existing methods including modified Yule-Walker and gradient descent and show our approach is, on average, both faster and more accurate.

Christian J. Steinmetz, Joshua D. Reiss

Deep learning approaches for beat and downbeat tracking have brought advancements. However, these approaches continue to rely on hand-crafted, subsampled spectral features as input, restricting the information available to the model. In this work, we propose WaveBeat, an end-to-end approach for joint beat and downbeat tracking operating directly on waveforms. This method forgoes engineered spectral features, and instead, produces beat and downbeat predictions directly from the waveform, the first of its kind for this task. Our model utilizes temporal convolutional networks (TCNs) operating on waveforms that achieve a very large receptive field ($\geq$ 30 s) at audio sample rates in a memory efficient manner by employing rapidly growing dilation factors with fewer layers. With a straightforward data augmentation strategy, our method outperforms previous state-of-the-art methods on some datasets, while producing comparable results on others, demonstrating the potential for time domain approaches.

Christian J. Steinmetz, Vamsi Krishna Ithapu, Paul Calamia

Deep learning approaches have emerged that aim to transform an audio signal so that it sounds as if it was recorded in the same room as a reference recording, with applications both in audio post-production and augmented reality. In this work, we propose FiNS, a Filtered Noise Shaping network that directly estimates the time domain room impulse response (RIR) from reverberant speech. Our domain-inspired architecture features a time domain encoder and a filtered noise shaping decoder that models the RIR as a summation of decaying filtered noise signals, along with direct sound and early reflection components. Previous methods for acoustic matching utilize either large models to transform audio to match the target room or predict parameters for algorithmic reverberators. Instead, blind estimation of the RIR enables efficient and realistic transformation with a single convolution. An evaluation demonstrates our model not only synthesizes RIRs that match parameters of the target room, such as the $T_{60}$ and DRR, but also more accurately reproduces perceptual characteristics of the target room, as shown in a listening test when compared to deep learning baselines.

Christian J. Steinmetz, Joshua D. Reiss

Deep learning approaches have demonstrated success in modeling analog audio effects. Nevertheless, challenges remain in modeling more complex effects that involve time-varying nonlinear elements, such as dynamic range compressors. Existing neural network approaches for modeling compression either ignore the device parameters, do not attain sufficient accuracy, or otherwise require large noncausal models prohibiting real-time operation. In this work, we propose a modification to temporal convolutional networks (TCNs) enabling greater efficiency without sacrificing performance. By utilizing very sparse convolutional kernels through rapidly growing dilations, our model attains a significant receptive field using fewer layers, reducing computation. Through a detailed evaluation we demonstrate our efficient and causal approach achieves state-of-the-art performance in modeling the analog LA-2A, is capable of real-time operation on CPU, and only requires 10 minutes of training data.

Christian J. Steinmetz, Jordi Pons, Santiago Pascual et al.

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 propose 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.

Christian J. Steinmetz, Joshua D. Reiss

By processing audio signals in the time-domain with randomly weighted temporal convolutional networks (TCNs), we uncover a wide range of novel, yet controllable overdrive effects. We discover that architectural aspects, such as the depth of the network, the kernel size, the number of channels, the activation function, as well as the weight initialization, all have a clear impact on the sonic character of the resultant effect, without the need for training. In practice, these effects range from conventional overdrive and distortion, to more extreme effects, as the receptive field grows, similar to a fusion of distortion, equalization, delay, and reverb. To enable use by musicians and producers, we provide a real-time plugin implementation. This allows users to dynamically design networks, listening to the results in real-time. We provide a demonstration and code at https://csteinmetz1.github.io/ronn.