Search

Multimodal foundation models, such as Gemini and ChatGPT, have revolutionized human-machine interactions by seamlessly integrating various forms of data. Developing a universal spoken language model that comprehends a wide range of natural language instructions is critical for bridging communication gaps and facilitating more intuitive interactions. However, the absence of a comprehensive evaluation benchmark poses a significant challenge. We present Dynamic-SUPERB Phase-2, an open and evolving benchmark for the comprehensive evaluation of instruction-based universal speech models. Building upon the first generation, this second version incorporates 125 new tasks contributed collaboratively by the global research community, expanding the benchmark to a total of 180 tasks, making it the largest benchmark for speech and audio evaluation. While the first generation of Dynamic-SUPERB was limited to classification tasks, Dynamic-SUPERB Phase-2 broadens its evaluation capabilities by introducing a wide array of novel and diverse tasks, including regression and sequence generation, across speech, music, and environmental audio. Evaluation results indicate that none of the models performed well universally. SALMONN-13B excelled in English ASR, while WavLLM demonstrated high accuracy in emotion recognition, but current models still require further innovations to handle a broader range of tasks. We will soon open-source all task data and the evaluation pipeline.

Audio-visual pre-trained models have gained substantial attention recently and demonstrated superior performance on various audio-visual tasks. This study investigates whether pre-trained audio-visual models demonstrate non-arbitrary associations between sounds and visual representations$\unicode{x2013}$known as sound symbolism$\unicode{x2013}$which is also observed in humans. We developed a specialized dataset with synthesized images and audio samples and assessed these models using a non-parametric approach in a zero-shot setting. Our findings reveal a significant correlation between the models' outputs and established patterns of sound symbolism, particularly in models trained on speech data. These results suggest that such models can capture sound-meaning connections akin to human language processing, providing insights into both cognitive architectures and machine learning strategies., Comment: Errors in the introduction part that might potentially affect the integrity of the paper. Withdraw at the point. Will replace with an updated version in the future

Clinical diagnosis of stuttering requires an assessment by a licensed speech-language pathologist. However, this process is time-consuming and requires clinicians with training and experience in stuttering and fluency disorders. Unfortunately, only a small percentage of speech-language pathologists report being comfortable working with individuals who stutter, which is inadequate to accommodate for the 80 million individuals who stutter worldwide. Developing machine learning models for detecting stuttered speech would enable universal and automated screening for stuttering, enabling speech pathologists to identify and follow up with patients who are most likely to be diagnosed with a stuttering speech disorder. Previous research in this area has predominantly focused on utterance-level detection, which is not sufficient for clinical settings where word-level annotation of stuttering is the norm. In this study, we curated a stuttered speech dataset with word-level annotations and introduced a word-level stuttering speech detection model leveraging self-supervised speech models. Our evaluation demonstrates that our model surpasses previous approaches in word-level stuttering speech detection. Additionally, we conducted an extensive ablation analysis of our method, providing insight into the most important aspects of adapting self-supervised speech models for stuttered speech detection., Comment: Accepted by IEEE SLT 2024

Self-supervised speech (SSL) models have recently become widely adopted for many downstream speech processing tasks. The general usage pattern is to employ SSL models as feature extractors, and then train a downstream prediction head to solve a specific task. However, different layers of SSL models have been shown to capture different types of information, and the methods of combining them are not well studied. To this end, we extend the general framework for SSL model utilization by proposing the interface that connects the upstream and downstream. Under this view, the dominant technique of combining features via a layerwise weighted sum can be regarded as a specific interface. We propose several alternative interface designs and demonstrate that the weighted sum interface is suboptimal for many tasks. In particular, we show that a convolutional interface whose depth scales logarithmically with the depth of the upstream model consistently outperforms many other interface designs., Comment: Accepted to Interspeech2024

The recently proposed visually grounded speech model SpeechCLIP is an innovative framework that bridges speech and text through images via CLIP without relying on text transcription. On this basis, this paper introduces two extensions to SpeechCLIP. First, we apply the Continuous Integrate-and-Fire (CIF) module to replace a fixed number of CLS tokens in the cascaded architecture. Second, we propose a new hybrid architecture that merges the cascaded and parallel architectures of SpeechCLIP into a multi-task learning framework. Our experimental evaluation is performed on the Flickr8k and SpokenCOCO datasets. The results show that in the speech keyword extraction task, the CIF-based cascaded SpeechCLIP model outperforms the previous cascaded SpeechCLIP model using a fixed number of CLS tokens. Furthermore, through our hybrid architecture, cascaded task learning boosts the performance of the parallel branch in image-speech retrieval tasks., Comment: Accepted to ICASSP 2024, Self-supervision in Audio, Speech, and Beyond (SASB) workshop

Recent advances in self-supervised speech models have shown significant improvement in many downstream tasks. However, these models predominantly centered on frame-level training objectives, which can fall short in spoken language understanding tasks that require semantic comprehension. Existing works often rely on additional speech-text data as intermediate targets, which is costly in the real-world setting. To address this challenge, we propose Pseudo-Word HuBERT (PW-HuBERT), a framework that integrates pseudo word-level targets into the training process, where the targets are derived from a visually-ground speech model, notably eliminating the need for speech-text paired data. Our experimental results on four spoken language understanding (SLU) benchmarks suggest the superiority of our model in capturing semantic information., Comment: Accepted to ICASSP 2024 workshop on Self-supervision in Audio, Speech, and Beyond (SASB)

The relationship between matrix phase composition and microstructure yielding in phase separated calcium aluminosilicate glasses is investigated. Varying the treatment temperature of a phase separated glass results in glasses with different microstructures and matrix compositions. The impact of matrix composition on hardness depends on the mode of microstructural deformation. In glasses that deform via the droplet-densification mechanism, decreasing the matrix silica content results in a decrease in hardness due to the weakest-link effect and greater amounts of incongruent yielding of the phases. This relationship may act as an additional source of the indentation size effect in phase separated glasses. In glasses that deform via droplet-coalescence, decreasing the matrix silica content results in increasing hardness due to increased coalescence of the droplet phase during yielding.

Audio-visual representation learning aims to develop systems with human-like perception by utilizing correlation between auditory and visual information. However, current models often focus on a limited set of tasks, and generalization abilities of learned representations are unclear. To this end, we propose the AV-SUPERB benchmark that enables general-purpose evaluation of unimodal audio/visual and bimodal fusion representations on 7 datasets covering 5 audio-visual tasks in speech and audio processing. We evaluate 5 recent self-supervised models and show that none of these models generalize to all tasks, emphasizing the need for future study on improving universal model performance. In addition, we show that representations may be improved with intermediate-task fine-tuning and audio event classification with AudioSet serves as a strong intermediate task. We release our benchmark with evaluation code and a model submission platform to encourage further research in audio-visual learning., Comment: Accepted to ICASSP 2024; Evaluation Code: https://github.com/roger-tseng/av-superb Submission Platform: https://av.superbbenchmark.org

Studies on artificial neural networks rarely address both vanishing gradients and overfitting issues. In this study, we follow the pupil learning procedure, which has the features of interpreting, picking, understanding, cramming, and organizing, to derive the pupil learning mechanism (PLM) by which to modify the network structure and weights of 2-layer neural networks (2LNNs). The PLM consists of modules for sequential learning, adaptive learning, perfect learning, and less-overfitted learning. Based upon a copper price forecasting dataset, we conduct an experiment to validate the PLM module design modules, and an experiment to evaluate the performance of PLM. The empirical results indeed approve the PLM module design and show the superiority of the proposed PLM model over the linear regression model and the conventional backpropagation-based 2LNN model.

This work investigates the use of large-scale, English-only pre-trained models (CLIP and HuBERT) for multilingual image-speech retrieval. For non-English image-speech retrieval, we outperform the current state-of-the-art performance by a wide margin both when training separate models for each language, and with a single model which processes speech in all three languages. We identify key differences in model behavior and performance between English and non-English settings, attributable to the English-only pre-training of CLIP and HuBERT, and investigate how fine-tuning the pre-trained models impacts these differences. Finally, we show that our models can be used for mono- and cross-lingual speech-text retrieval and cross-lingual speech-speech retrieval, despite never having seen any parallel speech-text or speech-speech data during training., Comment: Accepted to ICASSP 2023

Data-driven speech processing models usually perform well with a large amount of text supervision, but collecting transcribed speech data is costly. Therefore, we propose SpeechCLIP, a novel framework bridging speech and text through images to enhance speech models without transcriptions. We leverage state-of-the-art pre-trained HuBERT and CLIP, aligning them via paired images and spoken captions with minimal fine-tuning. SpeechCLIP outperforms prior state-of-the-art on image-speech retrieval and performs zero-shot speech-text retrieval without direct supervision from transcriptions. Moreover, SpeechCLIP can directly retrieve semantically related keywords from speech., Comment: Accepted to IEEE SLT 2022

Although neural radiance fields (NeRF) have shown impressive advances for novel view synthesis, most methods typically require multiple input images of the same scene with accurate camera poses. In this work, we seek to substantially reduce the inputs to a single unposed image. Existing approaches condition on local image features to reconstruct a 3D object, but often render blurry predictions at viewpoints that are far away from the source view. To address this issue, we propose to leverage both the global and local features to form an expressive 3D representation. The global features are learned from a vision transformer, while the local features are extracted from a 2D convolutional network. To synthesize a novel view, we train a multilayer perceptron (MLP) network conditioned on the learned 3D representation to perform volume rendering. This novel 3D representation allows the network to reconstruct unseen regions without enforcing constraints like symmetry or canonical coordinate systems. Our method can render novel views from only a single input image and generalize across multiple object categories using a single model. Quantitative and qualitative evaluations demonstrate that the proposed method achieves state-of-the-art performance and renders richer details than existing approaches., Comment: WACV 2023 Project website: https://cseweb.ucsd.edu/~viscomp/projects/VisionNeRF/

Objective: To develop a rule-based algorithm that detects temporal information of clinical events during pregnancy for women with COVID-19 by inferring gestational weeks and delivery dates from Electronic Health Records (EHR) from the National COVID Cohort Collaborate (N3C). Materials and Methods: The EHR are normalized by the Observational Medical Outcomes Partnership (OMOP) Clinical Data Model (CDM). EHR phenotyping resulted in 270,897 pregnant women (2018-06-01 to 2021-05-31). We developed a rule-based algorithm and performed a multi-level evaluation to test content validity and clinical validity of the algorithm; and extreme value analysis for individuals with <150 or >300 days of gestation. Results: The algorithm identified 296,194 pregnancies (16,659 COVID-19 174 and 744 without COVID-19 peri-pandemic) in 270,897 pregnant women. For inferring gestational age, 95% cases (n=40) have moderate-high accuracy (Cohen Kappa = 0.62); 100% cases (n=40) have moderate-high granularity of temporal information (Cohen Kappa = 1). For inferring delivery dates, the accuracy is 100% (Cohen Kappa = 1). Accuracy of gestational age detection for extreme length of gestation is 93.3% (Cohen Kappa = 1). Mothers with COVID-19 showed higher prevalence in obesity (35.1% vs. 29.5%), diabetes (17.8% vs. 17.0%), chronic obstructive pulmonary disease (COPD) (0.2% vs. 0.1%), respiratory distress syndrome (ARDS) (1.8% vs. 0.2%). Discussion: We explored the characteristics of pregnant women by different timing of COVID-19 with our algorithm: the first to infer temporal information from complete antenatal care and detect the timing of SARS-CoV-2 infection for pregnant women using N3C. Conclusion: The algorithm shows excellent validity in inferring gestational age and delivery dates, which supports national EHR cohorts on N3C studying the impact of COVID-19 on pregnancy.

In this paper, we introduce a shallow (one-hidden-layer) physics-informed neural network for solving partial differential equations on static and evolving surfaces. For the static surface case, with the aid of level set function, the surface normal and mean curvature used in the surface differential expressions can be computed easily. So instead of imposing the normal extension constraints used in literature, we write the surface differential operators in the form of traditional Cartesian differential operators and use them in the loss function directly. We perform a series of performance study for the present methodology by solving Laplace-Beltrami equation and surface diffusion equation on complex static surfaces. With just a moderate number of neurons used in the hidden layer, we are able to attain satisfactory prediction results. Then we extend the present methodology to solve the advection-diffusion equation on an evolving surface with given velocity. To track the surface, we additionally introduce a prescribed hidden layer to enforce the topological structure of the surface and use the network to learn the homeomorphism between the surface and the prescribed topology. The proposed network structure is designed to track the surface and solve the equation simultaneously. Again, the numerical results show comparable accuracy as the static cases. As an application, we simulate the surfactant transport on the droplet surface under shear flow and obtain some physically plausible results.

Attention-based Transformer models have been increasingly employed for automatic music generation. To condition the generation process of such a model with a user-specified sequence, a popular approach is to take that conditioning sequence as a priming sequence and ask a Transformer decoder to generate a continuation. However, this prompt-based conditioning cannot guarantee that the conditioning sequence would develop or even simply repeat itself in the generated continuation. In this paper, we propose an alternative conditioning approach, called theme-based conditioning, that explicitly trains the Transformer to treat the conditioning sequence as a thematic material that has to manifest itself multiple times in its generation result. This is achieved with two main technical contributions. First, we propose a deep learning-based approach that uses contrastive representation learning and clustering to automatically retrieve thematic materials from music pieces in the training data. Second, we propose a novel gated parallel attention module to be used in a sequence-to-sequence (seq2seq) encoder/decoder architecture to more effectively account for a given conditioning thematic material in the generation process of the Transformer decoder. We report on objective and subjective evaluations of variants of the proposed Theme Transformer and the conventional prompt-based baseline, showing that our best model can generate, to some extent, polyphonic pop piano music with repetition and plausible variations of a given condition., Comment: to be published at IEEE Transactions on Multimedia