Masked Modeling Duo: Towards a Universal Audio Pre-training Framework

Self-supervised learning (SSL) using masked prediction has made great strides in general-purpose audio representation. This study proposes Masked Modeling Duo (M2D), an improved masked prediction SSL, which learns by predicting representations of masked input signals that serve as training signals. Unlike conventional methods, M2D obtains a training signal by encoding only the masked part, encouraging the two networks in M2D to model the input. While M2D improves general-purpose audio representations, a specialized representation is essential for real-world applications, such as in industrial and medical domains. The often confidential and proprietary data in such domains is typically limited in size and has a different distribution from that in pre-training datasets. Therefore, we propose M2D for X (M2D-X), which extends M2D to enable the pre-training of specialized representations for an application X. M2D-X learns from M2D and an additional task and inputs background noise. We make the additional task configurable to serve diverse applications, while the background noise helps learn on small data and forms a denoising task that makes representation robust. With these design choices, M2D-X should learn a representation specialized to serve various application needs. Our experiments confirmed that the representations for general-purpose audio, specialized for the highly competitive AudioSet and speech domain, and a small-data medical task achieve top-level performance, demonstrating the potential of using our models as a universal audio pre-training framework. Our code is available online for future studies at https://github.com/nttcslab/m2d

Deep Attentive Time Warping

Similarity measures for time series are important problems for time series classification. To handle the nonlinear time distortions, Dynamic Time Warping (DTW) has been widely used. However, DTW is not learnable and suffers from a trade-off between robustness against time distortion and discriminative power. In this paper, we propose a neural network model for task-adaptive time warping. Specifically, we use the attention model, called the bipartite attention model, to develop an explicit time warping mechanism with greater distortion invariance. Unlike other learnable models using DTW for warping, our model predicts all local correspondences between two time series and is trained based on metric learning, which enables it to learn the optimal data-dependent warping for the target task. We also propose to induce pre-training of our model by DTW to improve the discriminative power. Extensive experiments demonstrate the superior effectiveness of our model over DTW and its state-of-the-art performance in online signature verification.

Audio Difference Captioning Utilizing Similarity-Discrepancy Disentanglement

We proposed Audio Difference Captioning (ADC) as a new extension task of audio captioning for describing the semantic differences between input pairs of similar but slightly different audio clips. The ADC solves the problem that conventional audio captioning sometimes generates similar captions for similar audio clips, failing to describe the difference in content. We also propose a cross-attention-concentrated transformer encoder to extract differences by comparing a pair of audio clips and a similarity-discrepancy disentanglement to emphasize the difference in the latent space. To evaluate the proposed methods, we built an AudioDiffCaps dataset consisting of pairs of similar but slightly different audio clips with human-annotated descriptions of their differences. The experiment with the AudioDiffCaps dataset showed that the proposed methods solve the ADC task effectively and improve the attention weights to extract the difference by visualizing them in the transformer encoder.

Masked Modeling Duo for Speech: Specializing General-Purpose Audio Representation to Speech using Denoising Distillation

Self-supervised learning general-purpose audio representations have demonstrated high performance in a variety of tasks. Although they can be optimized for application by fine-tuning, even higher performance can be expected if they can be specialized to pre-train for an application. This paper explores the challenges and solutions in specializing general-purpose audio representations for a specific application using speech, a highly demanding field, as an example. We enhance Masked Modeling Duo (M2D), a general-purpose model, to close the performance gap with state-of-the-art (SOTA) speech models. To do so, we propose a new task, denoising distillation, to learn from fine-grained clustered features, and M2D for Speech (M2D-S), which jointly learns the denoising distillation task and M2D masked prediction task. Experimental results show that M2D-S performs comparably to or outperforms SOTA speech models on the SUPERB benchmark, demonstrating that M2D can specialize in a demanding field. Our code is available at: https://github.com/nttcslab/m2d/tree/master/speech

Masked Modeling Duo: Learning Representations by Encouraging Both Networks to Model the Input

Masked Autoencoders is a simple yet powerful self-supervised learning method. However, it learns representations indirectly by reconstructing masked input patches. Several methods learn representations directly by predicting representations of masked patches; however, we think using all patches to encode training signal representations is suboptimal. We propose a new method, Masked Modeling Duo (M2D), that learns representations directly while obtaining training signals using only masked patches. In the M2D, the online network encodes visible patches and predicts masked patch representations, and the target network, a momentum encoder, encodes masked patches. To better predict target representations, the online network should model the input well, while the target network should also model it well to agree with online predictions. Then the learned representations should better model the input. We validated the M2D by learning general-purpose audio representations, and M2D set new state-of-the-art performance on tasks such as UrbanSound8K, VoxCeleb1, AudioSet20K, GTZAN, and SpeechCommandsV2.

Reflectance-Oriented Probabilistic Equalization for Image Enhancement

Despite recent advances in image enhancement, it remains difficult for existing approaches to adaptively improve the brightness and contrast for both low-light and normal-light images. To solve this problem, we propose a novel 2D histogram equalization approach. It assumes intensity occurrence and co-occurrence to be dependent on each other and derives the distribution of intensity occurrence (1D histogram) by marginalizing over the distribution of intensity co-occurrence (2D histogram). This scheme improves global contrast more effectively and reduces noise amplification. The 2D histogram is defined by incorporating the local pixel value differences in image reflectance into the density estimation to alleviate the adverse effects of dark lighting conditions. Over 500 images were used for evaluation, demonstrating the superiority of our approach over existing studies. It can sufficiently improve the brightness of low-light images while avoiding over-enhancement in normal-light images.

Reflectance-Guided, Contrast-Accumulated Histogram Equalization

Existing image enhancement methods fall short of expectations because with them it is difficult to improve global and local image contrast simultaneously. To address this problem, we propose a histogram equalization-based method that adapts to the data-dependent requirements of brightness enhancement and improves the visibility of details without losing the global contrast. This method incorporates the spatial information provided by image context in density estimation for discriminative histogram equalization. To minimize the adverse effect of non-uniform illumination, we propose defining spatial information on the basis of image reflectance estimated with edge preserving smoothing. Our method works particularly well for determining how the background brightness should be adaptively adjusted and for revealing useful image details hidden in the dark.

ConceptBeam: Concept Driven Target Speech Extraction

We propose a novel framework for target speech extraction based on semantic information, called ConceptBeam. Target speech extraction means extracting the speech of a target speaker in a mixture. Typical approaches have been exploiting properties of audio signals, such as harmonic structure and direction of arrival. In contrast, ConceptBeam tackles the problem with semantic clues. Specifically, we extract the speech of speakers speaking about a concept, i.e., a topic of interest, using a concept specifier such as an image or speech. Solving this novel problem would open the door to innovative applications such as listening systems that focus on a particular topic discussed in a conversation. Unlike keywords, concepts are abstract notions, making it challenging to directly represent a target concept. In our scheme, a concept is encoded as a semantic embedding by mapping the concept specifier to a shared embedding space. This modality-independent space can be built by means of deep metric learning using paired data consisting of images and their spoken captions. We use it to bridge modality-dependent information, i.e., the speech segments in the mixture, and the specified, modality-independent concept. As a proof of our scheme, we performed experiments using a set of images associated with spoken captions. That is, we generated speech mixtures from these spoken captions and used the images or speech signals as the concept specifiers. We then extracted the target speech using the acoustic characteristics of the identified segments. We compare ConceptBeam with two methods: one based on keywords obtained from recognition systems and another based on sound source separation. We show that ConceptBeam clearly outperforms the baseline methods and effectively extracts speech based on the semantic representation.

Introducing Auxiliary Text Query-modifier to Content-based Audio Retrieval

The amount of audio data available on public websites is growing rapidly, and an efficient mechanism for accessing the desired data is necessary. We propose a content-based audio retrieval method that can retrieve a target audio that is similar to but slightly different from the query audio by introducing auxiliary textual information which describes the difference between the query and target audio. While the range of conventional content-based audio retrieval is limited to audio that is similar to the query audio, the proposed method can adjust the retrieval range by adding an embedding of the auxiliary text query-modifier to the embedding of the query sample audio in a shared latent space. To evaluate our method, we built a dataset comprising two different audio clips and the text that describes the difference. The experimental results show that the proposed method retrieves the paired audio more accurately than the baseline. We also confirmed based on visualization that the proposed method obtains the shared latent space in which the audio difference and the corresponding text are represented as similar embedding vectors.

Composing General Audio Representation by Fusing Multilayer Features of a Pre-trained Model

Many application studies rely on audio DNN models pre-trained on a large-scale dataset as essential feature extractors, and they extract features from the last layers. In this study, we focus on our finding that the middle layer features of existing supervised pre-trained models are more effective than the late layer features for some tasks. We propose a simple approach to compose features effective for general-purpose applications, consisting of two steps: (1) calculating feature vectors along the time frame from middle/late layer outputs, and (2) fusing them. This approach improves the utility of frequency and channel information in downstream processes, and combines the effectiveness of middle and late layer features for different tasks. As a result, the feature vectors become effective for general purposes. In the experiments using VGGish, PANNs' CNN14, and AST on nine downstream tasks, we first show that each layer output of these models serves different tasks. Then, we demonstrate that the proposed approach significantly improves their performance and brings it to a level comparable to that of the state-of-the-art. In particular, the performance of the non-semantic speech (NOSS) tasks greatly improves, especially on Speech commands V2 with VGGish of +77.1 (14.3% to 91.4%).