Kohei Yatabe

Yuma Koizumi, Heiga Zen, Shigeki Karita et al.

Speech restoration (SR) is a task of converting degraded speech signals into high-quality ones. In this study, we propose a robust SR model called Miipher, and apply Miipher to a new SR application: increasing the amount of high-quality training data for speech generation by converting speech samples collected from the Web to studio-quality. To make our SR model robust against various degradation, we use (i) a speech representation extracted from w2v-BERT for the input feature, and (ii) a text representation extracted from transcripts via PnG-BERT as a linguistic conditioning feature. Experiments show that Miipher (i) is robust against various audio degradation and (ii) enable us to train a high-quality text-to-speech (TTS) model from restored speech samples collected from the Web. Audio samples are available at our demo page: google.github.io/df-conformer/miipher/

Yuma Koizumi, Kohei Yatabe, Heiga Zen et al.

Denoising diffusion probabilistic models (DDPMs) and generative adversarial networks (GANs) are popular generative models for neural vocoders. The DDPMs and GANs can be characterized by the iterative denoising framework and adversarial training, respectively. This study proposes a fast and high-quality neural vocoder called \textit{WaveFit}, which integrates the essence of GANs into a DDPM-like iterative framework based on fixed-point iteration. WaveFit iteratively denoises an input signal, and trains a deep neural network (DNN) for minimizing an adversarial loss calculated from intermediate outputs at all iterations. Subjective (side-by-side) listening tests showed no statistically significant differences in naturalness between human natural speech and those synthesized by WaveFit with five iterations. Furthermore, the inference speed of WaveFit was more than 240 times faster than WaveRNN. Audio demos are available at \url{google.github.io/df-conformer/wavefit/}.

Hideki Kawahara, Kohei Yatabe, Ken-Ichi Sakakibara et al.

We propose an objective measurement method for pitch extractors' responses to frequency-modulated signals. The method simultaneously measures the linear and the non-linear time-invariant responses and random and time-varying responses. It uses extended time-stretched pulses combined by binary orthogonal sequences. Our recent finding of involuntary voice pitch response to auditory stimulation while voicing motivated this proposal. The involuntary voice pitch response provides means to investigate voice chain subsystems individually and objectively. This response analysis requires reliable and precise pitch extraction. We found that existing pitch extractors failed to correctly analyze signals used for auditory stimulation by using the proposed method. Therefore, we propose two reference pitch extractors based on the instantaneous frequency analysis and multi-resolution power spectrum analysis. The proposed extractors correctly analyze the test signals. We open-sourced MATLAB codes to measure pitch extractors and codes for conducting the voice pitch response experiment on our GitHub repository.

Hideki Kawahara, Toshie Matsui, Kohei Yatabe et al.

Auditory feedback plays an essential role in the regulation of the fundamental frequency of voiced sounds. The fundamental frequency also responds to auditory stimulation other than the speaker's voice. We propose to use this response of the fundamental frequency of sustained vowels to frequency-modulated test signals for investigating involuntary control of voice pitch. This involuntary response is difficult to identify and isolate by the conventional paradigm, which uses step-shaped pitch perturbation. We recently developed a versatile measurement method using a mixture of orthogonal sequences made from a set of extended time-stretched pulses (TSP). In this article, we extended our approach and designed a set of test signals using the mixture to modulate the fundamental frequency of artificial signals. For testing the response, the experimenter presents the modulated signal aurally while the subject is voicing sustained vowels. We developed a tool for conducting this test quickly and interactively. We make the tool available as an open-source and also provide executable GUI-based applications. Preliminary tests revealed that the proposed method consistently provides compensatory responses with about 100 ms latency, representing involuntary control. Finally, we discuss future applications of the proposed method for objective and non-invasive auditory response measurements.

Hideki Kawahara, Kohei Yatabe

We introduce a new member of TSP (Time Stretched Pulse) for acoustic and speech measurement infrastructure, based on a simple all-pass filter and systematic randomization. This new infrastructure fundamentally upgrades our previous measurement procedure, which enables simultaneous measurement of multiple attributes, including non-linear ones without requiring extra filtering nor post-processing. Our new proposal establishes a theoretically solid, flexible, and extensible foundation in acoustic measurement. Moreover, it is general enough to provide versatile research tools for other fields, such as biological signal analysis. We illustrate using acoustic measurements and data augmentation as representative examples among various prospective applications. We open-sourced MATLAB implementation. It consists of an interactive and real-time acoustic tool, MATLAB functions, and supporting materials.

Kazuki Matsumoto, Ren Uchida, Kohei Yatabe

The robustness of deep neural networks (DNNs) can be certified through their Lipschitz continuity, which has made the construction of Lipschitz-continuous DNNs an active research field. However, DNNs for audio processing have not been a major focus due to their poor compatibility with existing results. In this paper, we consider the amplitude modifier (AM), a popular architecture for handling audio signals, and propose its Lipschitz-continuous variants, which we refer to as LipsAM. We prove a sufficient condition for an AM to be Lipschitz continuous and propose two architectures as examples of LipsAM. The proposed architectures were applied to a Plug-and-Play algorithm for speech dereverberation, and their improved stability is demonstrated through numerical experiments.

Tomohiko Nakamura, Kohei Yatabe

This paper proposes a universal sound separation (USS) method capable of handling untrained sampling frequencies (SFs). The USS aims at separating arbitrary sources of different types and can be the key technique to realize a source separator that can be universally used as a preprocessor for any downstream tasks. To realize a universal source separator, there are two essential properties: universalities with respect to source types and recording conditions. The former property has been studied in the USS literature, which has greatly increased the number of source types that can be handled by a single neural network. However, the latter property (e.g., SF) has received less attention despite its necessity. Since the SF varies widely depending on the downstream tasks, the universal source separator must handle a wide variety of SFs. In this paper, to encompass the two properties, we propose an SF-independent (SFI) extension of a computationally efficient USS network, SuDoRM-RF. The proposed network uses our previously proposed SFI convolutional layers, which can handle various SFs by generating convolutional kernels in accordance with an input SF. Experiments show that signal resampling can degrade the USS performance and the proposed method works more consistently than signal-resampling-based methods for various SFs.

Kazuaki Tanaka, Kohei Yatabe

The numerical solution of differential equations using neural networks has become a central topic in scientific computing, with Physics-Informed Neural Networks (PINNs) emerging as a powerful paradigm for both forward and inverse problems. However, unlike classical numerical methods that offer established convergence guarantees, neural network-based approximations typically lack rigorous error bounds. Furthermore, the non-deterministic nature of their optimization makes it difficult to mathematically certify their accuracy. To address these challenges, we propose a "Learn and Verify" framework that provides computable, mathematically rigorous error bounds for the solutions of differential equations. By combining a novel Doubly Smoothed Maximum (DSM) loss for training with interval arithmetic for verification, we compute rigorous a posteriori error bounds as machine-verifiable proofs. Numerical experiments on nonlinear Ordinary Differential Equations (ODEs), including problems with time-varying coefficients and finite-time blow-up, demonstrate that the proposed framework successfully constructs rigorous enclosures of the true solutions, establishing a foundation for trustworthy scientific machine learning.

Yuma Koizumi, Heiga Zen, Kohei Yatabe et al.

Neural vocoder using denoising diffusion probabilistic model (DDPM) has been improved by adaptation of the diffusion noise distribution to given acoustic features. In this study, we propose SpecGrad that adapts the diffusion noise so that its time-varying spectral envelope becomes close to the conditioning log-mel spectrogram. This adaptation by time-varying filtering improves the sound quality especially in the high-frequency bands. It is processed in the time-frequency domain to keep the computational cost almost the same as the conventional DDPM-based neural vocoders. Experimental results showed that SpecGrad generates higher-fidelity speech waveform than conventional DDPM-based neural vocoders in both analysis-synthesis and speech enhancement scenarios. Audio demos are available at wavegrad.github.io/specgrad/.

Kento Nagatomo, Masahiro Yasuda, Kohei Yatabe et al.

Sound event localization and detection (SELD) is a combined task of identifying the sound event and its direction. Deep neural networks (DNNs) are utilized to associate them with the sound signals observed by a microphone array. Although ambisonic microphones are popular in the literature of SELD, they might limit the range of applications due to their predetermined geometry. Some applications (including those for pedestrians that perform SELD while walking) require a wearable microphone array whose geometry can be designed to suit the task. In this paper, for the development of such a wearable SELD, we propose a dataset named Wearable SELD dataset. It consists of data recorded by 24 microphones placed on a head and torso simulators (HATS) with some accessories mimicking wearable devices (glasses, earphones, and headphones). We also provide experimental results of SELD using the proposed dataset and SELDNet to investigate the effect of microphone configuration.

Tomoro Tanaka, Kohei Yatabe, Masahiro Yasuda et al.

In this paper, we propose an audio declipping method that takes advantages of both sparse optimization and deep learning. Since sparsity-based audio declipping methods have been developed upon constrained optimization, they are adjustable and well-studied in theory. However, they always uniformly promote sparsity and ignore the individual properties of a signal. Deep neural network (DNN)-based methods can learn the properties of target signals and use them for audio declipping. Still, they cannot perform well if the training data have mismatches and/or constraints in the time domain are not imposed. In the proposed method, we use a DNN in an optimization algorithm. It is inspired by an idea called plug-and-play (PnP) and enables us to promote sparsity based on the learned information of data, considering constraints in the time domain. Our experiments confirmed that the proposed method is stable and robust to mismatches between training and test data.

Hideki Kawahara, Kohei Yatabe

We propose a simple method to measure acoustic responses using any sounds by converting them suitable for measurement. This method enables us to use music pieces for measuring acoustic conditions. It is advantageous to measure such conditions without annoying test sounds to listeners. In addition, applying the underlying idea of simultaneous measurement of multiple paths provides practically valuable features. For example, it is possible to measure deviations (temporally stable, random, and time-varying) and the impulse response while reproducing slightly modified contents under target conditions. The key idea of the proposed method is to add relatively small deterministic signals that sound like noise to the original sounds. We call the converted sounds safeguarded test signals.

Koichi Saito, Tomohiko Nakamura, Kohei Yatabe et al.

Audio source separation is often used as preprocessing of various applications, and one of its ultimate goals is to construct a single versatile model capable of dealing with the varieties of audio signals. Since sampling frequency, one of the audio signal varieties, is usually application specific, the preceding audio source separation model should be able to deal with audio signals of all sampling frequencies specified in the target applications. However, conventional models based on deep neural networks (DNNs) are trained only at the sampling frequency specified by the training data, and there are no guarantees that they work with unseen sampling frequencies. In this paper, we propose a convolution layer capable of handling arbitrary sampling frequencies by a single DNN. Through music source separation experiments, we show that the introduction of the proposed layer enables a conventional audio source separation model to consistently work with even unseen sampling frequencies.

Yoshiki Masuyama, Yoshiaki Bando, Kohei Yatabe et al.

Detecting sound source objects within visual observation is important for autonomous robots to comprehend surrounding environments. Since sounding objects have a large variety with different appearances in our living environments, labeling all sounding objects is impossible in practice. This calls for self-supervised learning which does not require manual labeling. Most of conventional self-supervised learning uses monaural audio signals and images and cannot distinguish sound source objects having similar appearances due to poor spatial information in audio signals. To solve this problem, this paper presents a self-supervised training method using 360° images and multichannel audio signals. By incorporating with the spatial information in multichannel audio signals, our method trains deep neural networks (DNNs) to distinguish multiple sound source objects. Our system for localizing sound source objects in the image is composed of audio and visual DNNs. The visual DNN is trained to localize sound source candidates within an input image. The audio DNN verifies whether each candidate actually produces sound or not. These DNNs are jointly trained in a self-supervised manner based on a probabilistic spatial audio model. Experimental results with simulated data showed that the DNNs trained by our method localized multiple speakers. We also demonstrate that the visual DNN detected objects including talking visitors and specific exhibits from real data recorded in a science museum.

Daichi Kitamura, Kohei Yatabe

Independent low-rank matrix analysis (ILRMA) is the state-of-the-art algorithm for blind source separation (BSS) in the determined situation (the number of microphones is greater than or equal to that of source signals). ILRMA achieves a great separation performance by modeling the power spectrograms of the source signals via the nonnegative matrix factorization (NMF). Such a highly developed source model can solve the permutation problem of the frequency-domain BSS to a large extent, which is the reason for the excellence of ILRMA. In this paper, we further improve the separation performance of ILRMA by additionally considering the general structure of spectrograms, which is called consistency, and hence we call the proposed method Consistent ILRMA. Since a spectrogram is calculated by an overlapping window (and a window function induces spectral smearing called main- and side-lobes), the time-frequency bins depend on each other. In other words, the time-frequency components are related to each other via the uncertainty principle. Such co-occurrence among the spectral components can function as an assistant for solving the permutation problem, which has been demonstrated by a recent study. On the basis of these facts, we propose an algorithm for realizing Consistent ILRMA by slightly modifying the original algorithm. Its performance was extensively evaluated through experiments performed with various window lengths and shift lengths. The results indicated several tendencies of the original and proposed ILRMA that include some topics not fully discussed in the literature. For example, the proposed Consistent ILRMA tends to outperform the original ILRMA when the window length is sufficiently long compared to the reverberation time of the mixing system.

Toru Nakashika, Kohei Yatabe

In audio applications, one of the most important representations of audio signals is the amplitude spectrogram. It is utilized in many machine-learning-based information processing methods including the ones using the restricted Boltzmann machines (RBM). However, the ordinary Gaussian-Bernoulli RBM (the most popular RBM among its variations) cannot directly handle amplitude spectra because the Gaussian distribution is a symmetric model allowing negative values which never appear in the amplitude. In this paper, after proposing a general gamma Boltzmann machine, we propose a practical model called the gamma-Bernoulli RBM that simultaneously handles both linear- and log-amplitude spectrograms. Its conditional distribution of the observable data is given by the gamma distribution, and thus the proposed RBM can naturally handle the data represented by positive numbers as the amplitude spectra. It can also treat amplitude in the logarithmic scale which is important for audio signals from the perceptual point of view. The advantage of the proposed model compared to the ordinary Gaussian-Bernoulli RBM was confirmed by PESQ and MSE in the experiment of representing the amplitude spectrograms of speech signals.

Kohei Yatabe

Multichannel audio blind source separation (BSS) in the determined situation (the number of microphones is equal to that of the sources), or determined BSS, is performed by multichannel linear filtering in the time-frequency domain to handle the convolutive mixing process. Ordinarily, the filter treats each frequency independently, which causes the well-known permutation problem, i.e., the problem of how to align the frequency-wise filters so that each separated component is correctly assigned to the corresponding sources. In this paper, it is shown that the general property of the time-frequency-domain representation called spectrogram consistency can be an assistant for solving the permutation problem.

Kohei Yatabe, Daichi Kitamura

This paper proposes harmonic vector analysis (HVA) based on a general algorithmic framework of audio blind source separation (BSS) that is also presented in this paper. BSS for a convolutive audio mixture is usually performed by multichannel linear filtering when the numbers of microphones and sources are equal (determined situation). This paper addresses such determined BSS based on batch processing. To estimate the demixing filters, effective modeling of the source signals is important. One successful example is independent vector analysis (IVA) that models the signals via co-occurrence among the frequency components in each source. To give more freedom to the source modeling, a general framework of determined BSS is presented in this paper. It is based on the plug-and-play scheme using a primal-dual splitting algorithm and enables us to model the source signals implicitly through a time-frequency mask. By using the proposed framework, determined BSS algorithms can be developed by designing masks that enhance the source signals. As an example of its application, we propose HVA by defining a time-frequency mask that enhances the harmonic structure of audio signals via sparsity of cepstrum. The experiments showed that HVA outperforms IVA and independent low-rank matrix analysis (ILRMA) for both speech and music signals. A MATLAB code is provided along with the paper for a reference ( https://doi.org/10.24433/CO.9507820.v1 ).

Masaki Kawanaka, Yuma Koizumi, Ryoichi Miyazaki et al.

Improving subjective sound quality of enhanced signals is one of the most important missions in speech enhancement. For evaluating the subjective quality, several methods related to perceptually-motivated objective sound quality assessment (OSQA) have been proposed such as PESQ (perceptual evaluation of speech quality). However, direct use of such measures for training deep neural network (DNN) is not allowed in most cases because popular OSQAs are non-differentiable with respect to DNN parameters. Therefore, the previous study has proposed to approximate the score of OSQAs by an auxiliary DNN so that its gradient can be used for training the primary DNN. One problem with this approach is instability of the training caused by the approximation error of the score. To overcome this problem, we propose to use stabilization techniques borrowed from reinforcement learning. The experiments, aimed to increase the score of PESQ as an example, show that the proposed method (i) can stably train a DNN to increase PESQ, (ii) achieved the state-of-the-art PESQ score on a public dataset, and (iii) resulted in better sound quality than conventional methods based on subjective evaluation.

Yuma Koizumi, Kohei Yatabe, Marc Delcroix et al.

This paper investigates a self-adaptation method for speech enhancement using auxiliary speaker-aware features; we extract a speaker representation used for adaptation directly from the test utterance. Conventional studies of deep neural network (DNN)--based speech enhancement mainly focus on building a speaker independent model. Meanwhile, in speech applications including speech recognition and synthesis, it is known that model adaptation to the target speaker improves the accuracy. Our research question is whether a DNN for speech enhancement can be adopted to unknown speakers without any auxiliary guidance signal in test-phase. To achieve this, we adopt multi-task learning of speech enhancement and speaker identification, and use the output of the final hidden layer of speaker identification branch as an auxiliary feature. In addition, we use multi-head self-attention for capturing long-term dependencies in the speech and noise. Experimental results on a public dataset show that our strategy achieves the state-of-the-art performance and also outperform conventional methods in terms of subjective quality.

Daiki Takeuchi, Kohei Yatabe, Yuma Koizumi et al.

We propose a speech enhancement method using a causal deep neural network~(DNN) for real-time applications. DNN has been widely used for estimating a time-frequency~(T-F) mask which enhances a speech signal. One popular DNN structure for that is a recurrent neural network~(RNN) owing to its capability of effectively modelling time-sequential data like speech. In particular, the long short-term memory (LSTM) is often used to alleviate the vanishing/exploding gradient problem which makes the training of an RNN difficult. However, the number of parameters of LSTM is increased as the price of mitigating the difficulty of training, which requires more computational resources. For real-time speech enhancement, it is preferable to use a smaller network without losing the performance. In this paper, we propose to use the equilibriated recurrent neural network~(ERNN) for avoiding the vanishing/exploding gradient problem without increasing the number of parameters. The proposed structure is causal, which requires only the information from the past, in order to apply it in real-time. Compared to the uni- and bi-directional LSTM networks, the proposed method achieved the similar performance with much fewer parameters.

Yoshiki Masuyama, Kohei Yatabe, Yuma Koizumi et al.

Phase reconstruction, which estimates phase from a given amplitude spectrogram, is an active research field in acoustical signal processing with many applications including audio synthesis. To take advantage of rich knowledge from data, several studies presented deep neural network (DNN)--based phase reconstruction methods. However, the training of a DNN for phase reconstruction is not an easy task because phase is sensitive to the shift of a waveform. To overcome this problem, we propose a DNN-based two-stage phase reconstruction method. In the proposed method, DNNs estimate phase derivatives instead of phase itself, which allows us to avoid the sensitivity problem. Then, phase is recursively estimated based on the estimated derivatives, which is named recurrent phase unwrapping (RPU). The experimental results confirm that the proposed method outperformed the direct phase estimation by a DNN.

Daiki Takeuchi, Kohei Yatabe, Yuma Koizumi et al.

We propose an end-to-end speech enhancement method with trainable time-frequency~(T-F) transform based on invertible deep neural network~(DNN). The resent development of speech enhancement is brought by using DNN. The ordinary DNN-based speech enhancement employs T-F transform, typically the short-time Fourier transform~(STFT), and estimates a T-F mask using DNN. On the other hand, some methods have considered end-to-end networks which directly estimate the enhanced signals without T-F transform. While end-to-end methods have shown promising results, they are black boxes and hard to understand. Therefore, some end-to-end methods used a DNN to learn the linear T-F transform which is much easier to understand. However, the learned transform may not have a property important for ordinary signal processing. In this paper, as the important property of the T-F transform, perfect reconstruction is considered. An invertible nonlinear T-F transform is constructed by DNNs and learned from data so that the obtained transform is perfectly reconstructing filterbank.

Daiki Takeuchi, Kohei Yatabe, Yuma Koizumi et al.

We propose a data-driven design method of perfect-reconstruction filterbank (PRFB) for sound-source enhancement (SSE) based on deep neural network (DNN). DNNs have been used to estimate a time-frequency (T-F) mask in the short-time Fourier transform (STFT) domain. Their training is more stable when a simple cost function as mean-squared error (MSE) is utilized comparing to some advanced cost such as objective sound quality assessments. However, such a simple cost function inherits strong assumptions on the statistics of the target and/or noise which is often not satisfied, and the mismatch of assumption results in degraded performance. In this paper, we propose to design the frequency scale of PRFB from training data so that the assumption on MSE is satisfied. For designing the frequency scale, the warped filterbank frame (WFBF) is considered as PRFB. The frequency characteristic of learned WFBF was in between STFT and the wavelet transform, and its effectiveness was confirmed by comparison with a standard STFT-based DNN whose input feature is compressed into the mel scale.

Yoshiki Masuyama, Kohei Yatabe, Yasuhiro Oikawa

Low-rankness of amplitude spectrograms has been effectively utilized in audio signal processing methods including non-negative matrix factorization. However, such methods have a fundamental limitation owing to their amplitude-only treatment where the phase of the observed signal is utilized for resynthesizing the estimated signal. In order to address this limitation, we directly treat a complex-valued spectrogram and show a complex-valued spectrogram of a sum of sinusoids can be approximately low-rank by modifying its phase. For evaluating the applicability of the proposed low-rank representation, we further propose a convex prior emphasizing harmonic signals, and it is applied to audio denoising.

Yoshiki Masuyama, Kohei Yatabe, Yasuhiro Oikawa

Decomposition of an audio mixture into harmonic and percussive components, namely harmonic/percussive source separation (HPSS), is a useful pre-processing tool for many audio applications. Popular approaches to HPSS exploit the distinctive source-specific structures of power spectrograms. However, such approaches consider only power spectrograms, and the phase remains intact for resynthesizing the separated signals. In this paper, we propose a phase-aware HPSS method based on the structure of the phase of harmonic components. It is formulated as a convex optimization problem in the time domain, which enables the simultaneous treatment of both amplitude and phase. The numerical experiment validates the effectiveness of the proposed method.

Yoshiki Masuyama, Kohei Yatabe, Yuma Koizumi et al.

This paper presents a novel phase reconstruction method (only from a given amplitude spectrogram) by combining a signal-processing-based approach and a deep neural network (DNN). To retrieve a time-domain signal from its amplitude spectrogram, the corresponding phase is required. One of the popular phase reconstruction methods is the Griffin-Lim algorithm (GLA), which is based on the redundancy of the short-time Fourier transform. However, GLA often involves many iterations and produces low-quality signals owing to the lack of prior knowledge of the target signal. In order to address these issues, in this study, we propose an architecture which stacks a sub-block including two GLA-inspired fixed layers and a DNN. The number of stacked sub-blocks is adjustable, and we can trade the performance and computational load based on requirements of applications. The effectiveness of the proposed method is investigated by reconstructing phases from amplitude spectrograms of speeches.

Tsubasa Kusano, Yoshiki Masuyama, Kohei Yatabe et al.

Many audio signal processing methods are formulated in the time-frequency (T-F) domain which is obtained by the short-time Fourier transform (STFT). The properties of the STFT are fully characterized by window function, number of frequency channels, and time-shift. Thus, designing a better window is important for improving the performance of the processing especially when a less redundant T-F representation is desirable. While many window functions have been proposed in the literature, they are designed to have a good frequency response for analysis, which may not perform well in terms of signal processing. The window design must take the effect of the reconstruction (from the T-F domain into the time domain) into account for improving the performance. In this paper, an optimization-based design method of a nearly tight window is proposed to obtain a window performing well for the T-F domain signal processing.