Conditional sound separation in multi-source audio mixtures without having access to single source sound data during training is a long standing challenge. Existing mix-and-separate based methods suffer from significant performance drop with multi-source training mixtures due to the lack of supervision signal for single source separation cases during training. However, in the case of language-conditional audio separation, we do have access to corresponding text descriptions for each audio mixture in our training data, which can be seen as (rough) representations of the audio samples in the language modality. To this end, in this paper, we propose a generic bi-modal separation framework which can enhance the existing unsupervised frameworks to separate single-source signals in a target modality (i.e., audio) using the easily separable corresponding signals in the conditioning modality (i.e., language), without having access to single-source samples in the target modality during training. We empirically show that this is well within reach if we have access to a pretrained joint embedding model between the two modalities (i.e., CLAP). Furthermore, we propose to incorporate our framework into two fundamental scenarios to enhance separation performance. First, we show that our proposed methodology significantly improves the performance of purely unsupervised baselines by reducing the distribution shift between training and test samples. In particular, we show that our framework can achieve 71% boost in terms of Signal-to-Distortion Ratio (SDR) over the baseline, reaching 97.5% of the supervised learning performance. Second, we show that we can further improve the performance of the supervised learning itself by 17% if we augment it by our proposed weakly-supervised framework, that enables a powerful semi-supervised framework for audio separation.
Masked Autoencoders (MAEs) learn rich low-level representations from unlabeled data but require substantial labeled data to effectively adapt to downstream tasks. Conversely, Instance Discrimination (ID) emphasizes high-level semantics, offering a potential solution to alleviate annotation requirements in MAEs. Although combining these two approaches can address downstream tasks with limited labeled data, naively integrating ID into MAEs leads to extended training times and high computational costs. To address this challenge, we introduce uaMix-MAE, an efficient ID tuning strategy that leverages unsupervised audio mixtures. Utilizing contrastive tuning, uaMix-MAE aligns the representations of pretrained MAEs, thereby facilitating effective adaptation to task-specific semantics. To optimize the model with small amounts of unlabeled data, we propose an audio mixing technique that manipulates audio samples in both input and virtual label spaces. Experiments in low/few-shot settings demonstrate that \modelname achieves 4-6% accuracy improvements over various benchmarks when tuned with limited unlabeled data, such as AudioSet-20K. Code is available at https://github.com/PLAN-Lab/uamix-MAE
Learned image compression has gained widespread popularity for their efficiency in achieving ultra-low bit-rates. Yet, images containing substantial textual content, particularly screen-content images (SCI), often suffers from text distortion at such compressed levels. To address this, we propose to minimize a novel text logit loss designed to quantify the disparity in text between the original and reconstructed images, thereby improving the perceptual quality of the reconstructed text. Through rigorous experimentation across diverse datasets and employing state-of-the-art algorithms, our findings reveal significant enhancements in the quality of reconstructed text upon integration of the proposed loss function with appropriate weighting. Notably, we achieve a Bjontegaard delta (BD) rate of -32.64% for Character Error Rate (CER) and -28.03% for Word Error Rate (WER) on average by applying the text logit loss for two screenshot datasets. Additionally, we present quantitative metrics tailored for evaluating text quality in image compression tasks. Our findings underscore the efficacy and potential applicability of our proposed text logit loss function across various text-aware image compression contexts.
Generalization remains a major problem in supervised learning of single-channel speech enhancement. In this work, we propose learnable loss mixup (LLM), a simple and effortless training diagram, to improve the generalization of deep learning-based speech enhancement models. Loss mixup, of which learnable loss mixup is a special variant, optimizes a mixture of the loss functions of random sample pairs to train a model on virtual training data constructed from these pairs of samples. In learnable loss mixup, by conditioning on the mixed data, the loss functions are mixed using a non-linear mixing function automatically learned via neural parameterization. Our experimental results on the VCTK benchmark show that learnable loss mixup achieves 3.26 PESQ, outperforming the state-of-the-art.
Speech disfluencies, such as filled pauses or repetitions, are disruptions in the typical flow of speech. Stuttering is a speech disorder characterized by a high rate of disfluencies, but all individuals speak with some disfluencies and the rates of disfluencies may by increased by factors such as cognitive load. Clinically, automatic disfluency detection may help in treatment planning for individuals who stutter. Outside of the clinic, automatic disfluency detection may serve as a pre-processing step to improve natural language understanding in downstream applications. With this wide range of applications in mind, we investigate language, acoustic, and multimodal methods for frame-level automatic disfluency detection and categorization. Each of these methods relies on audio as an input. First, we evaluate several automatic speech recognition (ASR) systems in terms of their ability to transcribe disfluencies, measured using disfluency error rates. We then use these ASR transcripts as input to a language-based disfluency detection model. We find that disfluency detection performance is largely limited by the quality of transcripts and alignments. We find that an acoustic-based approach that does not require transcription as an intermediate step outperforms the ASR language approach. Finally, we present multimodal architectures which we find improve disfluency detection performance over the unimodal approaches. Ultimately, this work introduces novel approaches for automatic frame-level disfluency and categorization. In the long term, this will help researchers incorporate automatic disfluency detection into a range of applications.
Diffusion models power a vast majority of text-to-audio (TTA) generation methods. Unfortunately, these models suffer from slow inference speed due to iterative queries to the underlying denoising network, thus unsuitable for scenarios with inference time or computational constraints. This work modifies the recently proposed consistency distillation framework to train TTA models that require only a single neural network query. In addition to incorporating classifier-free guidance into the distillation process, we leverage the availability of generated audio during distillation training to fine-tune the consistency TTA model with novel loss functions in the audio space, such as the CLAP score. Our objective and subjective evaluation results on the AudioCaps dataset show that consistency models retain diffusion models' high generation quality and diversity while reducing the number of queries by a factor of 400.
We study the problem of progressive distillation: Given a large, pre-trained teacher model $g$, we seek to decompose the model into an ensemble of smaller, low-inference cost student models $f_i$. The resulting ensemble allows for flexibly tuning accuracy vs. inference cost, which is useful for a number of applications in on-device inference. The method we propose, B-DISTIL, relies on an algorithmic procedure that uses function composition over intermediate activations to construct expressive ensembles with similar performance as $g$, but with much smaller student models. We demonstrate the effectiveness of \algA by decomposing pretrained models across standard image, speech, and sensor datasets. We also provide theoretical guarantees for our method in terms of convergence and generalization.
In recent years, Generative Adversarial Networks (GANs) have produced significantly improved results in speech enhancement (SE) tasks. They are difficult to train, however. In this work, we introduce several improvements to the GAN training schemes, which can be applied to most GAN-based SE models. We propose using consistency loss functions, which target the inconsistency in time and time-frequency domains caused by Fourier and Inverse Fourier Transforms. We also present self-correcting optimization for training a GAN discriminator on SE tasks, which helps avoid "harmful" training directions for parts of the discriminator loss function. We have tested our proposed methods on several state-of-the-art GAN-based SE models and obtained consistent improvements, including new state-of-the-art results for the Voice Bank+DEMAND dataset.
Improving generalization is a major challenge in audio classification due to labeled data scarcity. Self-supervised learning (SSL) methods tackle this by leveraging unlabeled data to learn useful features for downstream classification tasks. In this work, we propose an augmented contrastive SSL framework to learn invariant representations from unlabeled data. Our method applies various perturbations to the unlabeled input data and utilizes contrastive learning to learn representations robust to such perturbations. Experimental results on the Audioset and DESED datasets show that our framework significantly outperforms state-of-the-art SSL and supervised learning methods on sound/event classification tasks.
Deep learning-based speech enhancement has shown unprecedented performance in recent years. The most popular mono speech enhancement frameworks are end-to-end networks mapping the noisy mixture into an estimate of the clean speech. With growing computational power and availability of multichannel microphone recordings, prior works have aimed to incorporate spatial statistics along with spectral information to boost up performance. Despite an improvement in enhancement performance of mono output, the spatial image preservation and subjective evaluations have not gained much attention in the literature. This paper proposes a novel stereo-aware framework for speech enhancement, i.e., a training loss for deep learning-based speech enhancement to preserve the spatial image while enhancing the stereo mixture. The proposed framework is model independent, hence it can be applied to any deep learning based architecture. We provide an extensive objective and subjective evaluation of the trained models through a listening test. We show that by regularizing for an image preservation loss, the overall performance is improved, and the stereo aspect of the speech is better preserved.