Search

Abstract Self-supervised learning for speech pre-training models has achieved remarkable success in acquiring superior speech contextual representations by learning from unlabeled audio, excelling in numerous downstream speech tasks. However, the pre-training of these models necessitates significant computational resources and training duration, presenting a high barrier to entry into the realm of pre-training learning. In our efforts, by amalgamating the resource-efficient benefits of the generative learning model, Masked Auto Encoder, with the efficacy of the vector quantization method in discriminative learning, we introduce a novel pre-training framework: Speech Vector Quantization Masked Auto Encoder (SVQ-MAE). Distinct from the majority of SSL frameworks, which require simultaneous construction of speech contextual representations and mask reconstruction within an encoder-only module, we have exclusively designed a decoupled decoder for pre-training SVQ-MAE. This allows the additional decoupled decoder to undertake the mask reconstruction task solely, reducing the learning complexity of pretext tasks and enhancing the encoder’s efficiency in extracting speech contextual representations. Owing to this innovation, by using only 4 GPUs, SVQ-NAE can achieve high performance comparable to wav2vec 2.0, which requires 64 GPUs for training. In the Speech Processing Universal Performance Benchmark, SVQ-MAE surpasses wav2vec 2.0 in both keyword spotting and emotion recognition tasks. Furthermore, in cross-lingual ASR for Mandarin, upon fine-tuning on AISHELL-1, SVQ-MAE achieves a Character Error Rate of 4.09%, outperforming all supervised ASR models.

Abstract Binaural audio recordings become increasingly popular in multimedia repositories, posing new challenges in indexing, searching, and retrieval of such excerpts in terms of their spatial audio scene characteristics. This paper presents a new method for the automatic estimation of one of the most important spatial attributes of binaural recordings of music, namely “ensemble width.” The method has been developed using a repository of 23,040 binaural excerpts synthesized by convolving 192 multi-track music recordings with 30 sets of head-related transfer functions (HRTF). The synthesized excerpts represented various spatial distributions of music sound sources along a frontal semicircle in the horizontal plane. A binaural auditory model was exploited to derive the standard binaural cues from the synthesized excerpts, yielding a dataset representing interaural level and time differences, complemented by interaural cross-correlation coefficients. Subsequently, a regression method, based on gradient-boosted decision trees, was applied to the formerly calculated dataset to estimate ensemble width values. According to the obtained results, the mean absolute error of the ensemble width estimation averaged across experimental conditions amounts to 6.63° (SD 0.12°). The accuracy of the method is the highest for the recordings with ensembles narrower than 30°, yielding the mean absolute error ranging between 0.8° and 10.2°. The performance of the proposed algorithm is relatively uniform regardless of the horizontal position of an ensemble. However, its accuracy deteriorates for wider ensembles, with the error reaching 25.2° for the music ensembles spanning 90°. The developed method exhibits satisfactory generalization properties when evaluated both under music-independent and HRTF-independent conditions. The proposed method outperforms the technique based on “spatiograms” recently introduced in the literature.

Abstract Over the past decade, the prevalence of neurological diseases has significantly risen due to population growth and aging. Individuals suffering from spastic paralysis, brain attack, and idiopathic Parkinson’s disease (PD), among other neurological illnesses, commonly suffer from dysarthria. Early detection and treatment of dysarthria in these patients are essential for effectively managing the progression of their disease. This paper provides UTrans-DSR, a novel encoder-decoder architecture for analyzing Mel-spectrograms (generated from audios) and classifying speech as healthy or dysarthric. Our model employs transformer encoder features based on a hybrid design, which includes the feature enhancement block (FEB) and the vision transformer (ViT) encoders. This combination effectively extracts global and local pixel information regarding localization while optimizing the mel-spectrograms feature extraction process. We keep up with the original class-token grouping sequence in the vision transformer while generating a new equivalent expanding route. More specifically, two unique growing pathways use a deep-supervision approach to increase spatial data recovery and expedite model convergence. We add consecutive residual connections to the system to reduce feature loss while increasing spatial data retrieval. Our technique is based on identifying gaps in mel-spectrograms distinguishing between normal and dysarthric speech. We conducted several experiments on UTrans-DSR using the UA speech and TORGO datasets, and it outperformed the existing top models. The model performed significantly in pixel’s localized and spatial feature extraction, effectively detecting and classifying spectral gaps. The Tran-DSR model outperforms previous research models, achieving an accuracy of 97.75%.

Abstract This paper proposes novel methods for extracting a single Speech signal of Interest (SOI) from a multichannel observed signal in underdetermined situations, i.e., when the observed signal contains more speech signals than microphones. It focuses on extracting the SOI using prior knowledge of the SOI’s Direction of Arrival (DOA). Conventional beamformers (BFs) and Blind Source Separation (BSS) with spatial regularization struggle to suppress interference speech signals in such situations. Although Switching Minimum Power Distortionless Response BF (Sw-MPDR) can handle underdetermined situations using a switching mechanism, its estimation accuracy significantly decreases when it relies on a steering vector determined by the SOI’s DOA. Spatially-Regularized Independent Vector Extraction (SRIVE) can robustly enhance the SOI based solely on its DOA using spatial regularization, but its performance degrades in underdetermined situations. This paper extends these conventional methods to overcome their limitations. First, we introduce a time-varying Gaussian (TVG) source model to Sw-MPDR to effectively enhance the SOI based solely on the DOA. Second, we introduce the switching mechanism to SRIVE to improve its speech enhancement performance in underdetermined situations. These two proposed methods are called Switching weighted MPDR (Sw-wMPDR) and Switching SRIVE (Sw-SRIVE). We experimentally demonstrate that both surpass conventional methods in enhancing the SOI using the DOA in underdetermined situations.

Abstract Over the past few decades, extensive research has been devoted to the design of artificial reverberation algorithms aimed at emulating the room acoustics of physical environments. Despite significant advancements, automatic parameter tuning of delay-network models remains an open challenge. We introduce a novel method for finding the parameters of a feedback delay network (FDN) such that its output renders target attributes of a measured room impulse response. The proposed approach involves the implementation of a differentiable FDN with trainable delay lines, which, for the first time, allows us to simultaneously learn each and every delay-network parameter via backpropagation. The iterative optimization process seeks to minimize a perceptually motivated time-domain loss function incorporating differentiable terms accounting for energy decay and echo density. Through experimental validation, we show that the proposed method yields time-invariant frequency-independent FDNs capable of closely matching the desired acoustical characteristics and outperforms existing methods based on genetic algorithms and analytical FDN design.

Abstract This paper addresses the challenge of online blind speaker separation in a multi-microphone setting. The linearly constrained minimum variance (LCMV) beamformer is selected as the backbone of the separation algorithm due to its distortionless response and capacity to create a null towards interfering sources. A specific instance of the LCMV beamformer that considers acoustic propagation is implemented. In this variant, the relative transfer functions (RTFs) associated with each speaker of interest are utilized as the steering vectors of the beamformer. A control mechanism is devised to ensure robust estimation of the beamformer’s building blocks, comprising speaker activity detectors and direction of arrival (DOA) estimation branches. This control mechanism is implemented as a multi-task deep neural network (DNN). The primary task classifies each time frame based on speaker activity: no active speaker, single active speaker, or multiple active speakers. The secondary task is DOA estimation. It is implemented as a classification task, executed only for frames classified as single-speaker frames by the primary branch. The direction of the active speaker is classified into one of the multiple ranges of angles. These frames are also leveraged to estimate the RTFs using subspace estimation methods. A library of RTFs associated with these DOA ranges is then constructed, facilitating rapid acquisition of new speakers and efficient tracking of existing speakers. The proposed scheme is evaluated in both simulated and real-life recordings, encompassing static and dynamic scenarios. The benefits of the multi-task approach are showcased, and significant improvements are evident, even when the control mechanism is trained with simulated data and tested with real-life data. A comparison between the proposed scheme and the independent low-rank matrix analysis (ILRMA) algorithm reveals significant improvements in static scenarios. Furthermore, the tracking capabilities of the proposed scheme are highlighted in dynamic scenarios.

Abstract The sound produced by vehicles driving on roadways constitutes one of the dominant noise sources in urban areas. The impact of traffic noise on human activities and the related investigation on modeling, assessment, and abatement strategies fueled the research on the simulation of the sound produced by individual passing vehicles. Simulators enable in fact to promote a perceptual assessment of the nature of traffic noise and of the impact of single road agents on the overall soundscape. In this work, we present TrafficSoundSim, an open-source framework for the acoustic simulation of vehicles transiting on a road. We first discuss the generation of the sound signal produced by a vehicle, represented as a combination of road/tire interaction noise and engine noise. We then introduce a propagation model based on the use of variable length delay lines, allowing to simulate acoustic propagation and Doppler effect. The proposed simulator incorporates the effect of air absorption and ground reflection, modeled via complex-valued reflection coefficients dependent on the road surface impedance, as well as a model of the directivity of sound sources representing the passing vehicles. The source signal generation and the propagation stages are decoupled, and all effects are implemented using finite impulse response filters. Moreover, no recorded data is required to run the simulation, making the framework flexible and independent on data availability. Finally, to validate the framework capability to accurately simulate passing vehicles, a comparison between synthetic and recorded pass-by events is presented. The validation shows that sounds generated with the proposed method achieve a good match with recorded events in terms of power spectral density and psychoacoustics metrics as well as a perceptually plausible result.

Abstract With the rapid development of social networks, short videos have become a popular form of content, especially dance videos. In this context, research on automatically generating music for dance videos shows significant practical value. However, existing studies face challenges such as limited richness in music timbre and lack of synchronization with dance movements. In this paper, we propose Dance2Music-Diffusion, a novel framework for music generation from dance videos using latent diffusion models. Our approach includes a motion encoder module for extracting motion features and a music diffusion generation module for generating latent music representations. By integrating dance type monitoring and latent diffusion techniques, our framework outperforms existing methods in generating complex and rich dance music. We conducted objective and subjective evaluations of the results produced by various existing models on the AIST++ dataset. Our framework shows outstanding performance in terms of beat recall rate, consistency with GT beats, and coordination with dance movements. This work represents the state of the art in automatic music generation from dance videos, is easy to train, and has implications for enhancing entertainment experiences and inspiring innovative dance productions. Sample videos of our generated music and dance can be viewed at https://youtu.be/eCvLdLdkX-Y . The code is available at https://github.com/hellto/dance2music-diffusion .

Abstract Multi-channel speech enhancement plays a critical role in numerous speech-related applications. Several previous works explicitly utilize deep neural networks (DNNs) to exploit tempo-spectral signal characteristics, which often leads to excellent performance. In this work, we present a time-frequency fusion model, namely TFFM, for multi-channel speech enhancement. We utilize three cascaded U-Nets to capture three types of high-resolution features, aiming to investigate their individual contributions. To be specific, the first U-Net keeps the time dimension and performs feature extraction along the frequency dimension for the high-resolution spectral features with global temporal information, the second U-Net keeps the frequency dimension and extracts features along the time dimension for the high-resolution temporal features with global spectral information, and the third U-Net downsamples and upsamples along both the frequency and time dimensions for the high-resolution tempo-spectral features. These three cascaded U-Nets are designed to aggregate local and global features, thereby effectively handling the tempo-spectral information of speech signals. The proposed TFFM in this work outperforms state-of-the-art baselines.

Abstract We propose a framework for classifying acoustic scenes utilizing distributed sound sensor devices capable of sound-to-light conversion, which we term as Blinkies. These Blinkies can convert acoustic signals into varying intensities of light via an inbuilt light-emitting diode. By using Blinkies, we can aggregate the spatial acoustic information across a wide region by recording the fluctuating light intensities of numerous Blinkies distributed throughout the region. Nonetheless, the signal communicated is subject to the bandwidth limitation imposed by the frame rate of the video camera, typically capped at 30 frames per second. Our objective is to refine the process of transforming sound into light for the purpose of acoustic scene classification within these bandwidth confines. While traversing through the air, a light signal is affected by inherent physical limitations such as the attenuation of light and interference from noise. To account for these factors, we have integrated these physical constraints into differentiable physical layers. This approach enables us to jointly train a pair of deep neural networks for the conversion of sound to light and for the classification of acoustic scenes. Our simulation studies, which employed the SINS database for acoustic scene classification, demonstrated that our proposed framework outperforms the previous one that utilized Blinkies. These findings emphasize the effectiveness of Blinkies in the field of acoustic scene classification.

Abstract Nowadays, the application of artificial intelligence (AI) algorithms and techniques is ubiquitous and transversal. Fields that take advantage of AI advances include sound and music processing. The advances in interdisciplinary research potentially yield new insights that may further advance the AI methods in this field. This special issue aims to report recent progress and spur new research lines in AI-driven sound and music processing, especially within interdisciplinary research scenarios.

Abstract Recent developments in acoustic signal processing have seen the integration of deep learning methodologies, alongside the continued prominence of classical wave expansion-based approaches, particularly in sound field reconstruction. Physics-informed neural networks (PINNs) have emerged as a novel framework, bridging the gap between data-driven and model-based techniques for addressing physical phenomena governed by partial differential equations. This paper introduces a PINN-based approach for the recovery of arbitrary volumetric acoustic fields. The network incorporates the wave equation to impose a regularization on signal reconstruction in the time domain. This methodology enables the network to learn the physical law of sound propagation and allows for the complete characterization of the sound field based on a limited set of observations. The proposed method’s efficacy is validated through experiments involving speech signals in a real-world environment, considering varying numbers of available measurements. Moreover, a comparative analysis is undertaken against state-of-the-art frequency domain and time domain reconstruction methods from existing literature, highlighting the increased accuracy across the various measurement configurations.

Abstract Room impulse responses (RIRs) are used in several applications, such as augmented reality and virtual reality. These applications require a large number of RIRs to be convolved with audio, under strict latency constraints. In this paper, we consider the compression of RIRs, in conjunction with fast time-domain convolution. We consider three different methods of RIR approximation for the purpose of RIR compression and compare them to state-of-the-art compression. The methods are evaluated using several standard objective quality measures, both channel-based and signal-based. We also propose a novel low-rank-based algorithm for fast time-domain convolution and show how the convolution can be carried out without the need to decompress the RIR. Numerical simulations are performed using RIRs of different lengths, recorded in three different rooms. It is shown that compression using low-rank approximation is a very compelling option to the state-of-the-art Opus compression, as it performs as well or better than on all but one considered measure, with the added benefit of being amenable to fast time-domain convolution.

Abstract A kernel interpolation method for the acoustic transfer function (ATF) between regions constrained by the physics of sound while being adaptive to the data is proposed. Most ATF interpolation methods aim to model the ATF for fixed source by using techniques that fit the estimation to the measurements while not taking the physics of the problem into consideration. We aim to interpolate the ATF for a region-to-region estimation, meaning we account for variation of both source and receiver positions. By using a very general formulation for the reproducing kernel function, we have created a kernel function that considers both directed and residual fields as two separate kernel functions. The directed field kernel considers a sparse selection of reflective field components with large amplitudes and is formulated as a combination of directional kernels. The residual field is composed of the remaining densely distributed components with lower amplitudes. Its kernel weight is represented by a universal approximator, a neural network, in order to learn patterns from the data freely. These kernel parameters are learned using Bayesian inference both under the assumption of Gaussian priors and by using a Markov chain Monte Carlo simulation method to perform inference in a more directed manner. We compare all established kernel formulations with each other in numerical simulations, showing that the proposed kernel model is capable of properly representing the complexities of the ATF.

Abstract The estimation sound fields over space is of interest in sound field control and analysis, spatial audio, room acoustics and virtual reality. Sound fields can be estimated from a number of measurements distributed over space yet this remains a challenging problem due to the large experimental effort required. In this work we investigate sensor distributions that are optimal to estimate sound fields. Such optimization is valuable as it can greatly reduce the number of measurements required. The sensor positions are optimized with respect to the parameters describing a sound field, or the pressure reconstructed at the area of interest, by finding the positions that minimize the Bayesian Cramér-Rao bound (BCRB). The optimized distributions are investigated in a numerical study as well as with measured room impulse responses. We observe a reduction in the number of measurements of approximately 50% when the sensor positions are optimized for reconstructing the sound field when compared with random distributions. The results indicate that optimizing the sensors positions is also valuable when the vector of parameters is sparse, specially compared with random sensor distributions, which are often adopted in sparse array processing in acoustics.

Abstract This paper presents the crossing scheme (X-scheme) for improving the performance of deep neural network (DNN)-based music source separation (MSS) with almost no increasing calculation cost. It consists of three components: (i) multi-domain loss (MDL), (ii) bridging operation, which couples the individual instrument networks, and (iii) combination loss (CL). MDL enables the taking advantage of the frequency- and time-domain representations of audio signals. We modify the target network, i.e., the network architecture of the original DNN-based MSS, by adding bridging paths for each output instrument to share their information. MDL is then applied to the combinations of the output sources as well as each independent source; hence, we called it CL. MDL and CL can easily be applied to many DNN-based separation methods as they are merely loss functions that are only used during training and do not affect the inference step. Bridging operation does not increase the number of learnable parameters in the network. Experimental results showed that the validity of Open-Unmix (UMX), densely connected dilated DenseNet (D3Net) and convolutional time-domain audio separation network (Conv-TasNet) extended with our X-scheme, respectively called X-UMX, X-D3Net and X-Conv-TasNet, by comparing them with their original versions. We also verified the effectiveness of X-scheme in a large-scale data regime, showing its generality with respect to data size. X-UMX Large (X-UMXL), which was trained on large-scale internal data and used in our experiments, is newly available at https://github.com/asteroid-team/asteroid/tree/master/egs/musdb18/X-UMX .

Abstract End-to-end (E2E) automatic speech recognition (ASR) models, which consist of deep learning models, are able to perform ASR tasks using a single neural network. These models should be trained using a large amount of data; however, collecting speech data which matches the targeted speech domain can be difficult, so speech data is often used that is not an exact match to the target domain, resulting in lower performance. In comparison to speech data, in-domain text data is much easier to obtain. Thus, traditional ASR systems use separately trained language models and HMM-based acoustic models. However, it is difficult to separate language information from an E2E ASR model because the model learns both acoustic and language information in an integrated manner, making it very difficult to create E2E ASR models for specialized target domain which are able to achieve sufficient recognition performance at a reasonable cost. In this paper, we propose a method of replacing the language information within pre-trained E2E ASR models in order to achieve adaptation to a target domain. This is achieved by deleting the “implicit” language information contained within the ASR model by subtracting the source-domain language model trained with a transcription of the ASR’s training data in a logarithmic domain. We then integrate a target domain language model through addition in the logarithmic domain. This subtraction and addition to replace of the language model is based on Bayes’ theorem. In our experiment, we first used two datasets of the Corpus of Spontaneous Japanese (CSJ) to evaluate the effectiveness of our method. We then we evaluated our method using the Japanese Newspaper Article Speech (JNAS) and CSJ corpora, which contain audio data from the read speech and spontaneous speech domain, respectively, to test the effectiveness of our proposed method at bridging the gap between these two language domains. Our results show that our proposed language model replacement method achieved better ASR performance than both non-adapted (baseline) ASR models and ASR models adapted using the conventional Shallow Fusion method.

Abstract The disparities in phonetics and corpuses across the three major dialects of Tibetan exacerbate the difficulty of a single task model for one dialect to accommodate other different dialects. To address this issue, this paper proposes task-diverse meta-learning. Our model can acquire more comprehensive and robust features, facilitating its adaptation to the variations among different dialects. This study uses Tibetan dialect ID recognition and Tibetan speaker recognition as the source tasks for meta-learning, which aims to augment the ability of the model to discriminate variations and differences among different dialects. Consequently, the model’s performance in Tibetan multi-dialect speech recognition tasks is enhanced. The experimental results show that task-diverse meta-learning leads to improved performance in Tibetan multi-dialect speech recognition. This demonstrates the effectiveness and applicability of task-diverse meta-learning, thereby contributing to the advancement of speech recognition techniques in multi-dialect environments.

Abstract This paper introduces a simplified and controllable model for mode coupling in the context of modal synthesis. The model employs efficient coupled filters for sound synthesis purposes, intended to emulate the generation of sounds radiated by sources under strongly nonlinear conditions. Such filters generate tonal components in an interdependent way and are intended to emulate realistic perceptually salient effects in musical instruments in an efficient manner. The control of energy transfer between the filters is realized through a coupling matrix. The generation of prototypical sounds corresponding to nonlinear sources with the filter bank is presented. In particular, examples are proposed to generate sounds corresponding to impacts on thin structures and to the perturbation of the vibration of objects when it collides with an other object. The sound examples presented in the paper and available for listening on the accompanying site illustrate that a simple control of the input parameters allows the generation of sounds whose evocation is coherent and that the addition of random processes yields a significant improvement to the realism of the generated sounds.

Abstract End-to-end speech to text translation aims to directly translate speech from one language into text in another, posing a challenging cross-modal task particularly in scenarios of limited data. Multi-task learning serves as an effective strategy for knowledge sharing between speech translation and machine translation, which allows models to leverage extensive machine translation data to learn the mapping between source and target languages, thereby improving the performance of speech translation. However, in multi-task learning, finding a set of weights that balances various tasks is challenging and computationally expensive. We proposed an adaptive multi-task learning method to dynamically adjust multi-task weights based on the proportional losses incurred during training, enabling adaptive balance in multi-task learning for speech to text translation. Moreover, inherent representation disparities across different modalities impede speech translation models from harnessing textual data effectively. To bridge the gap across different modalities, we proposed to apply optimal transport in the input of end-to-end model to find the alignment between speech and text sequences and learn the shared representations between them. Experimental results show that our method effectively improved the performance on the Tibetan-Chinese, English-German, and English-French speech translation datasets.

Abstract Dysarthria is a speech disorder that affects the ability to communicate due to articulation difficulties. This research proposes a novel method for automatic dysarthria detection (ADD) and automatic dysarthria severity level assessment (ADSLA) by using a variable continuous wavelet transform (CWT) layered convolutional neural network (CNN) model. To determine their efficiency, the proposed model is assessed using two distinct corpora, TORGO and UA-Speech, comprising both dysarthria patients and healthy subject speech signals. The research study explores the effectiveness of CWT-layered CNN models that employ different wavelets such as Amor, Morse, and Bump. The study aims to analyze the models’ performance without the need for feature extraction, which could provide deeper insights into the effectiveness of the models in processing complex data. Also, raw waveform modeling preserves the original signal’s integrity and nuance, making it ideal for applications like speech recognition, signal processing, and image processing. Extensive analysis and experimentation have revealed that the Amor wavelet surpasses the Morse and Bump wavelets in accurately representing signal characteristics. The Amor wavelet outperforms the others in terms of signal reconstruction fidelity, noise suppression capabilities, and feature extraction accuracy. The proposed CWT-layered CNN model emphasizes the importance of selecting the appropriate wavelet for signal-processing tasks. The Amor wavelet is a reliable and precise choice for applications. The UA-Speech dataset is crucial for more accurate dysarthria classification. Advanced deep learning techniques can simplify early intervention measures and expedite the diagnosis process.

Abstract Polyphonic sound source localization and detection (SSLD) task aims to recognize the categories of sound events, identify their onset and offset times, and detect their corresponding direction-of-arrival (DOA), where polyphonic refers to the occurrence of multiple overlapping sound sources in a segment. However, vanilla SSLD methods based on convolutional recurrent neural network (CRNN) suffer from insufficient feature extraction. The convolutions with kernel of single scale in CRNN fail to adequately extract multi-scale features of sound events, which have diverse time-frequency characteristics. It results in that the extracted features lack fine-grained information helpful for the localization of sound sources. In response to these challenges, we propose a polyphonic SSLD network based on global-local feature extraction and recalibration (GLFER-Net), where the global-local feature (GLF) extractor is designed to extract the multi-scale global features through an omni-directional dynamic convolution (ODConv) layer and multi-scale feature extraction (MSFE) module. The local feature extraction (LFE) unit is designed for capturing detailed information. Besides, we design a feature recalibration (FR) module to emphasize the crucial features along multiple dimensions. On the open datasets of Task3 in DCASE 2021 and 2022 Challenges, we compared our proposed GLFER-Net with six and four SSLD methods, respectively. The results show that the GLFER-Net achieves competitive performance. The modules we designed are verified to be effective through a series of ablation experiments and visualization analyses.

Abstract While deep learning technologies have made remarkable progress in generating deepfakes, their misuse has become a well-known concern. As a result, the ubiquitous usage of deepfakes for increasing false information poses significant risks to the security and privacy of individuals. The primary objective of audio spoofing detection is to identify audio generated through numerous AI-based techniques. Several techniques for fake audio detection already exist using machine learning algorithms. However, they lack generalization and may not identify all types of AI-synthesized audios such as replay attacks, voice conversion, and text-to-speech (TTS). In this paper, a deep layered model, i.e., VGGish, along with an attention block, namely Convolutional Block Attention Module (CBAM) for spoofing detection, is introduced. Our suggested model successfully classifies input audio into two classes: Fake and Real, converting them into mel-spectrograms, and extracting their most representative features due to the attention block. Our model is a significant technique to utilize for audio spoofing detection due to a simple layered architecture. It captures complex relationships in audio signals due to both spatial and channel features present in an attention module. To evaluate the effectiveness of our model, we have conducted in-depth testing using the ASVspoof 2019 dataset. The proposed technique achieved an EER of 0.52% for Physical Access (PA) attacks and 0.07 % for Logical Access (LA) attacks.

Abstract A new method for estimating the first and second derivatives of discrete audio signals intended to achieve higher computational precision in analyzing the performance and characteristics of digital audio systems is presented. The method could find numerous applications in modeling nonlinear audio circuit systems, e.g., for audio synthesis and creating audio effects, music recognition and classification, time-frequency analysis based on nonstationary audio signal decomposition, audio steganalysis and digital audio authentication or audio feature extraction methods. The proposed algorithm employs the ordinary 7 point-stencil central-difference formulas with improvements that minimize the round-off and truncation errors. This is achieved by treating the step size of numerical differentiation as a regularization parameter, which acts as a decision threshold in all calculations. This approach requires shifting discrete audio data by fractions of the initial sample rate, which was obtained by fractional delay FIR filters designed with modified 11-term cosine-sum windows for interpolation and shifting of audio signals. The maximum relative error in estimating first and second derivatives of discrete audio signals are respectively in order of $$10^{-13}$$ 10 - 13 and $$10^{-10}$$ 10 - 10 over the entire audio band, which is close to double-precision floating-point accuracy for the first and better than single-precision floating-point accuracy for the second derivative estimation. Numerical testing showed that this performance of the proposed method is not influenced by the type of signal being differentiated (either stationary or nonstationary), and provides better results than other known differentiation methods, in the audio band up to 21 kHz.

Abstract This work introduces a large dataset comprising impulse responses of spatially distributed sources within a plane parallel to a planar microphone array. The dataset, named MIRACLE, encompasses 856,128 single-channel impulse responses and includes four different measurement scenarios. Three measurement scenarios were conducted under anechoic conditions. The fourth scenario includes an additional specular reflection from a reflective panel. The source positions were obtained by uniformly discretizing a rectangular source plane parallel to the microphone for each scenario. The dataset contains three scenarios with a spatial resolution of $$23\,\textrm{mm}$$ 23 mm at two different source-plane-to-array distances, as well as a scenario with a resolution of $$5\,\textrm{mm}$$ 5 mm for the shorter distance. In contrast to existing room impulse response datasets, the accuracy of the provided source location labels is assessed and additional metadata, such as the directivity of the loudspeaker used for excitation, is provided. The MIRACLE dataset can be used as a benchmark for data-driven modelling and interpolation methods as well as for various acoustic machine learning tasks, such as source separation, localization, and characterization. Two timely applications of the dataset are presented in this work: the generation of microphone array data for data-driven source localization and characterization tasks and data-driven model order reduction.

Abstract Time signature detection is a fundamental task in music information retrieval, aiding in music organization. In recent years, the demand for robust and efficient methods in music analysis has amplified, underscoring the significance of advancements in time signature detection. In this study, we explored the effectiveness of residual networks for time signature detection. Additionally, we compared the performance of the residual network (ResNet18) to already existing models such as audio similarity matrix (ASM) and beat similarity matrix (BSM). We also juxtaposed with traditional algorithms such as support vector machine (SVM), random forest, K-nearest neighbor (KNN), naive Bayes, and that of deep learning models, such as convolutional neural network (CNN) and convolutional recurrent neural network (CRNN). The evaluation is conducted using Mel-frequency cepstral coefficients (MFCCs) as feature representations on the Meter2800 dataset. Our results indicate that ResNet18 outperforms all other models thereby showing the potential of deep learning models for accurate time signature detection.

Abstract Limited data availability remains a significant challenge for Whisper’s low-resource speech recognition performance, falling short of practical application requirements. While previous studies have successfully reduced the recognition error rates of target language speech through fine-tuning, a comprehensive exploration and analysis of Whisper’s fine-tuning capabilities and the advantages and disadvantages of various fine-tuning strategies are still lacking. This paper aims to fill this gap by conducting comprehensive experimental exploration for Whisper’s low-resource speech recognition performance using five fine-tuning strategies with limited supervised data from seven low-resource languages. The results and analysis demonstrate that all fine-tuning strategies explored in this paper significantly enhance Whisper’s performance. However, different strategies vary in their suitability and practical effectiveness, highlighting the need for careful selection based on specific use cases and resources available.

Abstract In the era of advanced text-to-speech (TTS) systems capable of generating high-fidelity, human-like speech by referring a reference speech, voice cloning (VC), or zero-shot TTS (ZS-TTS), stands out as an important subtask. A primary challenge in VC is maintaining speech quality and speaker similarity with limited reference data for a specific speaker. However, existing VC systems often rely on naive combinations of embedded speaker vectors for speaker control, which compromises the capture of speaking style, voice print, and semantic accuracy. To overcome this, we introduce the Two-branch Speaker Control Module (TSCM), a novel and highly adaptable voice cloning module designed to precisely processing speaker or style control for a target speaker. Our method uses an advanced fusion of local-level features from a Gated Convolutional Network (GCN) and utterance-level features from a gated recurrent unit (GRU) to enhance speaker control. We demonstrate the effectiveness of TSCM by integrating it into advanced TTS systems like FastSpeech 2 and VITS architectures, significantly optimizing their performance. Experimental results show that TSCM enables accurate voice cloning for a target speaker with minimal data through both zero-shot or few-shot fine-tuning of pretrained TTS models. Furthermore, our TSCM-based VITS (TSCM-VITS) showcases superior performance in zero-shot scenarios compared to existing state-of-the-art VC systems, even with basic dataset configurations. Our method’s superiority is validated through comprehensive subjective and objective evaluations. A demonstration of our system is available at https://great-research.github.io/tsct-tts-demo/ , providing practical insights into its application and effectiveness.

Abstract Selective attention is a crucial ability of the auditory system. Computationally, following an auditory object can be illustrated as tracking its acoustic properties, e.g., pitch, timbre, or location in space. The difficulty is related to the fact that in a complex auditory scene, the information about the tracked object is not available in a clean form. The more cluttered the sound mixture, the more time and frequency regions where the object of interest is masked by other sound sources. How does the auditory system recognize and follow acoustic objects based on this fragmentary information? Numerous studies highlight the crucial role of top-down processing in this task. Having in mind both auditory modeling and signal processing applications, we investigated how computational methods with and without top-down processing deal with increasing sparsity of the auditory features in the task of estimating instantaneous voice states, defined as a combination of three parameters: fundamental frequency F0 and formant frequencies F1 and F2. We found that the benefit from top-down processing grows with increasing sparseness of the auditory data.

Abstract This work studies neural modeling of nonlinear parametric audio circuits, focusing on how the diversity of settings of the target device user controls seen during training affects network generalization. To study the problem, a large corpus of training datasets is synthetically generated using SPICE simulations of two distinct devices, an analog equalizer and an analog distortion pedal. A proven recurrent neural network architecture is trained using each dataset. The difference in the datasets is in the sampling resolution of the device user controls and in their overall size. Based on objective and subjective evaluation of the trained models, a sampling resolution of five for the device parameters is found to be sufficient to capture the behavior of the target systems for the types of devices considered during the study. This result is desirable, since a dense sampling grid can be impractical to realize in the general case when no automated way of setting the device parameters is available, while collecting large amounts of data using a sparse grid only incurs small additional costs. Thus, the result provides guidance for efficient collection of training data for neural modeling of other similar audio devices.

Abstract Visual speech recognition (VSR) is a challenging task that has received increasing interest during the last few decades. Current state of the art employs powerful end-to-end architectures based on deep learning which depend on large amounts of data and high computational resources for their estimation. We address the task of VSR for data scarcity scenarios with limited computational resources by using traditional approaches based on hidden Markov models. We present a novel learning strategy that employs information obtained from previous acoustic temporal alignments to improve the visual system performance. Furthermore, we studied multiple visual speech representations and how image resolution or frame rate affect its performance. All these experiments were conducted on the limited data VLRF corpus, a database which offers an audio-visual support to address continuous speech recognition in Spanish. The results show that our approach significantly outperforms the best results achieved on the task to date.

Abstract This article introduces Mi-Go, a tool aimed at evaluating the performance and adaptability of general-purpose speech recognition machine learning models across diverse real-world scenarios. The tool leverages YouTube as a rich and continuously updated data source, accounting for multiple languages, accents, dialects, speaking styles, and audio quality levels. To demonstrate the effectiveness of the tool, an experiment was conducted, by using Mi-Go to evaluate state-of-the-art automatic speech recognition machine learning models. The evaluation involved a total of 141 randomly selected YouTube videos. The results underscore the utility of YouTube as a valuable data source for evaluation of speech recognition models, ensuring their robustness, accuracy, and adaptability to diverse languages and acoustic conditions. Additionally, by contrasting the machine-generated transcriptions against human-made subtitles, the Mi-Go tool can help pinpoint potential misuse of YouTube subtitles, like search engine optimization.

Abstract Dynamic parameterization of acoustic environments has drawn widespread attention in the field of audio processing. Precise representation of local room acoustic characteristics is crucial when designing audio filters for various audio rendering applications. Key parameters in this context include reverberation time (RT $$_{60}$$ 60 ) and geometric room volume. In recent years, neural networks have been extensively applied in the task of blind room parameter estimation. However, there remains a question of whether pure attention mechanisms can achieve superior performance in this task. To address this issue, this study employs blind room parameter estimation based on monaural noisy speech signals. Various model architectures are investigated, including a proposed attention-based model. This model is a convolution-free Audio Spectrogram Transformer, utilizing patch splitting, attention mechanisms, and cross-modality transfer learning from a pretrained Vision Transformer. Experimental results suggest that the proposed attention mechanism-based model, relying purely on attention mechanisms without using convolution, exhibits significantly improved performance across various room parameter estimation tasks, especially with the help of dedicated pretraining and data augmentation schemes. Additionally, the model demonstrates more advantageous adaptability and robustness when handling variable-length audio inputs compared to existing methods.

Abstract In robotics, echolocation has been used to detect acoustic reflectors, e.g., walls, as it aids the robotic platform to navigate in darkness and also helps detect transparent surfaces. However, the transfer function or response of an acoustic system, e.g., loudspeakers/emitters, contributes to non-ideal behavior within the acoustic systems that can contribute to a phase lag due to propagation delay. This non-ideal response can hinder the performance of a time-of-arrival (TOA) estimator intended for acoustic reflector localization especially when the estimation of multiple reflections is required. In this paper, we, therefore, propose a robust expectation-maximization (EM) algorithm that takes into account the response of acoustic systems to enhance the TOA estimation accuracy when estimating multiple reflections when the robot is placed in a corner of a room. A non-ideal transfer function is built with two parameters, which are estimated recursively within the estimator. To test the proposed method, a hardware proof-of-concept setup was built with two different designs. The experimental results show that the proposed method could detect an acoustic reflector up to a distance of 1.6 m with $$60\%$$ 60 % accuracy under the signal-to-noise ratio (SNR) of 0 dB. Compared to the state-of-the-art EM algorithm, our proposed method provides improved performance when estimating TOA by $$10\%$$ 10 % under a low SNR value.

Abstract Speech signals are often distorted by reverberation and noise, with a widely distributed signal-to-noise ratio (SNR). To address this, our study develops robust, deep neural network (DNN)-based speech enhancement methods. We reproduce several DNN-based monaural speech enhancement methods and outline a strategy for constructing datasets. This strategy, validated through experimental reproductions, has effectively enhanced the denoising efficiency and robustness of the models. Then, we propose a causal speech enhancement system named Supervised Attention Multi-Scale Temporal Convolutional Network (SA-MSTCN). SA-MSTCN extracts the complex compressed spectrum (CCS) for input encoding and employs complex ratio masking (CRM) for output decoding. The supervised attention module, a lightweight addition to SA-MSTCN, guides feature extraction. Experiment results show that the supervised attention module effectively improves noise reduction performance with a minor increase in computational cost. The multi-scale temporal convolutional network refines the perceptual field and better reconstructs the speech signal. Overall, SA-MSTCN not only achieves state-of-the-art speech quality and intelligibility compared to other methods but also maintains stable denoising performance across various environments.

Abstract Spoofed speeches are becoming a big threat to society due to advancements in artificial intelligence techniques. Therefore, there must be an automated spoofing detector that can be integrated into automatic speaker verification (ASV) systems. In this study, we recommend a novel and robust model, named DeepDet, based on deep-layered architecture, to categorize speech into two classes: spoofed and bonafide. DeepDet is an improved model based on Yet Another Mobile Network (YAMNet) employing a customized MobileNet combined with a bottleneck attention module (BAM). First, we convert audio into mel-spectrograms that consist of time–frequency representations on mel-scale. Second, we trained our deep layered model using the extracted mel-spectrograms on a Logical Access (LA) set, including synthesized speeches and voice conversions of the ASVspoof-2019 dataset. In the end, we classified the audios, utilizing our trained binary classifier. More precisely, we utilized the power of layered architecture and guided attention that can discern the spoofed speech from bonafide samples. Our proposed improved model employs depth-wise linearly separate convolutions, which makes our model lighter weight than existing techniques. Furthermore, we implemented extensive experiments to assess the performance of the suggested model using the ASVspoof 2019 corpus. We attained an equal error rate (EER) of 0.042% on Logical Access (LA), whereas 0.43% on Physical Access (PA) attacks. Therefore, the performance of the proposed model is significant on the ASVspoof 2019 dataset and indicates the effectiveness of the DeepDet over existing spoofing detectors. Additionally, our proposed model is robust enough that can identify the unseen spoofed audios and classifies the several attacks accurately.

Abstract Technologies in healthcare, smart homes, security, ecology, and entertainment all deploy audio event detection (AED) in order to detect sound events in an audio recording. Effective AED techniques rely heavily on supervised or semi-supervised models to capture the wide range of dynamics spanned by sound events in order to achieve temporally precise boundaries and accurate event classification. These methods require extensive collections of labeled or weakly labeled in-domain data, which is costly and labor-intensive. Importantly, these approaches do not fully leverage the inherent variability and range of dynamics across sound events, aspects that can be effectively identified through unsupervised methods. The present work proposes an approach based on multi-rate autoencoders that are pretrained in an unsupervised way to leverage unlabeled audio data and ultimately learn the rich temporal dynamics inherent in natural sound events. This approach utilizes parallel autoencoders that achieve decompositions of the modulation spectrum along different bands. In addition, we introduce a rate-selective temporal contrastive loss to align the training objective with event detection metrics. Optimizing the configuration of multi-rate encoders and the temporal contrastive loss leads to notable improvements in domestic sound event detection in the context of the DCASE challenge.

Abstract Audio augmented reality (AAR), a prominent topic in the field of audio, requires understanding the listening environment of the user for rendering an authentic virtual auditory object. Reverberation time ( $$RT_{60}$$ R T 60 ) is a predominant metric for the characterization of room acoustics and numerous approaches have been proposed to estimate it blindly from a reverberant speech signal. However, a single $$RT_{60}$$ R T 60 value may not be sufficient to correctly describe and render the acoustics of a room. This contribution presents a method for the estimation of multiple room acoustic parameters required to render close-to-accurate room acoustics in an unknown environment. It is shown how these parameters can be estimated blindly using an audio transformer that can be deployed on a mobile device. Furthermore, the paper also discusses the use of the estimated room acoustic parameters to find a similar room from a dataset of real BRIRs that can be further used for rendering the virtual audio source. Additionally, a novel binaural room impulse response (BRIR) augmentation technique to overcome the limitation of inadequate data is proposed. Finally, the proposed method is validated perceptually by means of a listening test.

Abstract Most soundfield synthesis approaches deal with extensive and regular loudspeaker arrays, which are often not suitable for home audio systems, due to physical space constraints. In this article, we propose a technique for soundfield synthesis through more easily deployable irregular loudspeaker arrays, i.e., where the spacing between loudspeakers is not constant, based on deep learning. The input are the driving signals obtained through a plane wave decomposition-based technique. While the considered driving signals are able to correctly reproduce the soundfield with a regular array, they show degraded performances when using irregular setups. Through a complex-valued convolutional neural network (CNN), we modify the driving signals in order to compensate the errors in the reproduction of the desired soundfield. Since no ground truth driving signals are available for the compensated ones, we train the model by calculating the loss between the desired soundfield at a number of control points and the one obtained through the driving signals estimated by the network. The proposed model must be retrained for each irregular loudspeaker array configuration. Numerical results show better reproduction accuracy with respect to the plane wave decomposition-based technique, pressure-matching approach, and linear optimizers for driving signal compensation.

Abstract Audio effects are an ubiquitous tool in music production due to the interesting ways in which they can shape the sound of music. Guitar effects, the subset of all audio effects focusing on guitar signals, are commonly used in popular music to shape the guitar sound to fit specific genres or to create more variety within musical compositions. Automatic extraction of guitar effects and their parameter settings, with the aim to copy a target guitar sound, has been previously investigated, where artificial neural networks first determine the effect class of a reference signal and subsequently the parameter settings. These approaches require a corresponding guitar effect implementation to be available. In general, for very close sound matching, additional research regarding effect implementations is necessary. In this work, we present a different approach to circumvent these issues. We propose blind extraction of guitar effects through a combination of blind system inversion and neural guitar effect modeling. That way, an immediately usable, blind copy of the target guitar effect is obtained. The proposed method is tested with the phaser, softclipping and slapback delay effect. Listening tests with eight subjects indicate excellent quality of the blind copies, i.e., little to no difference to the reference guitar effect.

Abstract This work constitutes the first approach for automatically classifying the surface that the voiding flow impacts in non-invasive sound uroflowmetry tests using machine learning. Often, the voiding flow impacts the toilet walls (traditionally made of ceramic) instead of the water in the toilet. This may cause a reduction in the strength of the recorded audio signal, leading to a decrease in the amplitude of the extracted envelope. As a result, just from analysing the envelope, it is impossible to tell if that reduction in the envelope amplitude is due to a reduction in the voiding flow or an impact on the toilet wall. In this work, we study the classification of sound uroflowmetry data in male subjects depending on the surface that the urine impacts within the toilet: the three classes are water, ceramic and silence (where silence refers to an interruption of the voiding flow). We explore three frequency bands to study the feasibility of removing the human-speech band (below 8 kHz) to preserve user privacy. Regarding the classification task, three machine learning algorithms were evaluated: the support vector machine, random forest and k-nearest neighbours. These algorithms obtained accuracies of 96%, 99.46% and 99.05%, respectively. The algorithms were trained on a novel dataset consisting of audio signals recorded in four standard Spanish toilets. The dataset consists of 6481 1-s audio signals labelled as silence, voiding on ceramics and voiding on water. The obtained results represent a step forward in evaluating sound uroflowmetry tests without requiring patients to always aim the voiding flow at the water. We open the door for future studies that attempt to estimate the flow parameters and reconstruct the signal envelope based on the surface that the urine hits in the toilet.

Abstract The vast amount of information stored in audio repositories makes necessary the development of efficient and automatic methods to search on audio content. In that direction, search on speech (SoS) has received much attention in the last decades. To motivate the development of automatic systems, ALBAYZIN evaluations include a search on speech challenge since 2012. This challenge releases several databases that cover different acoustic domains (i.e., spontaneous speech from TV shows, conference talks, parliament sessions, to name a few) aiming to build automatic systems that retrieve a set of terms from those databases. This paper presents a baseline system based on the Whisper automatic speech recognizer for the spoken term detection task in the search on speech challenge held in 2022 within the ALBAYZIN evaluations. This baseline system will be released with this publication and will be given to participants in the upcoming SoS ALBAYZIN evaluation in 2024. Additionally, several analyses based on some term properties (i.e., in-language and foreign terms, and single-word and multi-word terms) are carried out to show the Whisper capability at retrieving terms that convey specific properties. Although the results obtained for some databases are far from being perfect (e.g., for broadcast news domain), this Whisper-based approach has obtained the best results on the challenge databases so far so that it presents a strong baseline system for the upcoming challenge, encouraging participants to improve it.

Abstract Analyzing songs is a problem that is being investigated to aid various operations on music access platforms. At the beginning of these problems is the identification of the person who sings the song. In this study, a singer identification application, which consists of Turkish singers and works for the Turkish language, is proposed in order to find a solution to this problem. Mel-spectrogram and octave-based spectral contrast values are extracted from the songs, and these values are combined into a hybrid feature vector. Thus, problem-specific situations such as determining the differences in the voices of the singers and reducing the effects of the year and album differences on the result are discussed. As a result of the tests and systematic evaluations, it has been shown that a certain level of success has been achieved in the determination of the singer who sings the song, and that the song is in a stable structure against the changes in the singing style and song structure. The results were analyzed in a database of 9 singers and 180 songs. An accuracy value of 89.4% was obtained using the reduction of the feature vector by PCA, the normalization of the data, and the Extra Trees classifier. Precision, recall and f-score values were 89.9%, 89.4% and 89.5%, respectively.

Abstract Claimed identities of speakers can be verified by means of automatic speaker verification (ASV) systems, also known as voice biometric systems. Focusing on security and robustness against spoofing attacks on ASV systems, and observing that the investigation of attacker’s perspectives is capable of leading the way to prevent known and unknown threats to ASV systems, several countermeasures (CMs) have been proposed during ASVspoof 2015, 2017, 2019, and 2021 challenge campaigns that were organized during INTERSPEECH conferences. Furthermore, there is a recent initiative to organize the ASVSpoof 5 challenge with the objective of collecting the massive spoofing/deepfake attack data (i.e., phase 1), and the design of a spoofing-aware ASV system using a single classifier for both ASV and CM, to design integrated CM-ASV solutions (phase 2). To that effect, this paper presents a survey on a diversity of possible strategies and vulnerabilities explored to successfully attack an ASV system, such as target selection, unavailability of global countermeasures to reduce the attacker’s chance to explore the weaknesses, state-of-the-art adversarial attacks based on machine learning, and deepfake generation. This paper also covers the possibility of attacks, such as hardware attacks on ASV systems. Finally, we also discuss the several technological challenges from the attacker’s perspective, which can be exploited to come up with better defence mechanisms for the security of ASV systems.

Abstract Chinese traditional music, a vital expression of Chinese cultural heritage, possesses both a profound emotional resonance and artistic allure. This study sets forth to refine and analyze the acoustical features essential for the aesthetic recognition of Chinese traditional music, utilizing a dataset spanning five aesthetic genres. Through recursive feature elimination, we distilled an initial set of 447 low-level physical features to a more manageable 44, establishing their feature-importance coefficients. This reduction allowed us to estimate the quantified influence of higher-level musical components on aesthetic recognition, following the establishment of a correlation between these components and their physical counterparts. We conducted a comprehensive examination of the impact of various musical elements on aesthetic genres. Our findings indicate that the selected 44-dimensional feature set could enhance aesthetic recognition. Among the high-level musical factors, timbre emerges as the most influential, followed by rhythm, pitch, and tonality. Timbre proved pivotal in distinguishing between the JiYang and BeiShang genres, while rhythm and tonality were key in differentiating LingDong from JiYang, as well as LingDong from BeiShang.

Abstract Recent advancements in deep learning-based speech enhancement models have extensively used attention mechanisms to achieve state-of-the-art methods by demonstrating their effectiveness. This paper proposes a transformer attention network based sub-convolutional U-Net (TANSCUNet) for speech enhancement. Instead of adopting conventional RNNs and temporal convolutional networks for sequence modeling, we employ a novel transformer-based attention network between the sub-convolutional U-Net encoder and decoder for better feature learning. More specifically, it is composed of several adaptive time―frequency attention modules and an adaptive hierarchical attention module, aiming to capture long-term time-frequency dependencies and further aggregate hierarchical contextual information. Additionally, a sub-convolutional encoder-decoder model used different kernel sizes to extract multi-scale local and contextual features from the noisy speech. The experimental results show that the proposed model outperforms several state-of-the-art methods.

Abstract Speech synthesis has made significant strides thanks to the transition from machine learning to deep learning models. Contemporary text-to-speech (TTS) models possess the capability to generate speech of exceptionally high quality, closely mimicking human speech. Nevertheless, given the wide array of applications now employing TTS models, mere high-quality speech generation is no longer sufficient. Present-day TTS models must also excel at producing expressive speech that can convey various speaking styles and emotions, akin to human speech. Consequently, researchers have concentrated their efforts on developing more efficient models for expressive speech synthesis in recent years. This paper presents a systematic review of the literature on expressive speech synthesis models published within the last 5 years, with a particular emphasis on approaches based on deep learning. We offer a comprehensive classification scheme for these models and provide concise descriptions of models falling into each category. Additionally, we summarize the principal challenges encountered in this research domain and outline the strategies employed to tackle these challenges as documented in the literature. In the Section 8, we pinpoint some research gaps in this field that necessitate further exploration. Our objective with this work is to give an all-encompassing overview of this hot research area to offer guidance to interested researchers and future endeavors in this field.

Abstract Accurately representing the sound field with high spatial resolution is crucial for immersive and interactive sound field reproduction technology. In recent studies, there has been a notable emphasis on efficiently estimating sound fields from a limited number of discrete observations. In particular, kernel-based methods using Gaussian processes (GPs) with a covariance function to model spatial correlations have been proposed. However, the current methods rely on pre-defined kernels for modeling, requiring the manual identification of optimal kernels and their parameters for different sound fields. In this work, we propose a novel approach that parameterizes GPs using a deep neural network based on neural processes (NPs) to reconstruct the magnitude of the sound field. This method has the advantage of dynamically learning kernels from data using an attention mechanism, allowing for greater flexibility and adaptability to the acoustic properties of the sound field. Numerical experiments demonstrate that our proposed approach outperforms current methods in reconstructing accuracy, providing a promising alternative for sound field reconstruction.

Abstract Shouted and normal speech classification plays an important role in many speech-related applications. The existing works are often based on magnitude-based features and ignore phase-based features, which are directly related to magnitude information. In this paper, the importance of phase-based features is explored for the detection of shouted speech. The novel contributions of this work are as follows. (1) Three phase-based features, namely, relative phase (RP), linear prediction analysis estimated speech-based RP (LPAES-RP) and linear prediction residual-based RP (LPR-RP) features, are explored for shouted and normal speech classification. (2) We propose a new RP feature, called the glottal source-based RP (GRP) feature. The main idea of the proposed GRP feature is to exploit the difference between RP and LPAES-RP features to detect shouted speech. (3) A score combination of phase- and magnitude-based features is also employed to further improve the classification performance. The proposed feature and combination are evaluated using the shouted normal electroglottograph speech (SNE-Speech) corpus. The experimental findings show that the RP, LPAES-RP, and LPR-RP features provide promising results for the detection of shouted speech. We also find that the proposed GRP feature can provide better results than those of the standard mel-frequency cepstral coefficient (MFCC) feature. Moreover, compared to using individual features, the score combination of the MFCC and RP/LPAES-RP/LPR-RP/GRP features yields an improved detection performance. Performance analysis under noisy environments shows that the score combination of the MFCC and the RP/LPAES-RP/LPR-RP features gives more robust classification. These outcomes show the importance of RP features in distinguishing shouted speech from normal speech.

Abstract Acoustic scene classification (ASC) is the process of identifying the acoustic environment or scene from which an audio signal is recorded. In this work, we propose an encoder-decoder-based approach to ASC, which is borrowed from the SegNet in image semantic segmentation tasks. We also propose a novel feature normalization method named Mixup Normalization, which combines channel-wise instance normalization and the Mixup method to learn useful information for scene and discard specific information related to different devices. In addition, we propose an event extraction block, which can extract the accurate semantic segmentation region from the segmentation network, to imitate the effect of image segmentation on audio features. With four data augmentation techniques, our best single system achieved an average accuracy of 71.26% on different devices in the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 ASC Task 1A dataset. The result indicates a minimum margin of 17% against the DCASE 2020 challenge Task 1A baseline system. It has lower complexity and higher performance compared with other state-of-the-art CNN models, without using any supplementary data other than the official challenge dataset.