Various parameter-efficient fine-tuning (PEFT) methods, including LoRA and its variants, have gained popularity for reducing computational costs. However, there is often an accuracy gap between PEFT approaches and full fine-tuning (FT), and this discrepancy has not yet been systematically explored. In this work, we introduce a method for selecting sparse sub-matrices that aims to minimize the performance gap between PEFT vs. full fine-tuning (FT) while also reducing both fine-tuning computational costs and memory costs. We explored both gradient-based and activation-based parameter selection methods to identify the most significant sub-matrices for downstream tasks, updating only these blocks during fine-tuning. In our experiments, we demonstrated that SMT consistently surpasses other PEFT baselines (e.g., LoRA and DoRA) in fine-tuning popular large language models such as LLaMA across a broad spectrum of tasks, while reducing the GPU memory footprint by 67% compared to FT. We also examine how the performance of LoRA and DoRA tends to plateau and decline as the number of trainable parameters increases, in contrast, our SMT method does not suffer from such issues.

Despite recent progress, machine learning (ML) models for open-domain audio generation need to catch up to generative models for image, text, speech, and music. The lack of massive open-domain audio datasets is the main reason for this performance gap; we overcome this challenge through a novel data augmentation approach. We leverage state-of-the-art (SOTA) Large Language Models (LLMs) to enrich captions in the weakly-labeled audio dataset. We then use a SOTA video-captioning model to generate captions for the videos from which the audio data originated, and we again use LLMs to merge the audio and video captions to form a rich, large-scale dataset. We experimentally evaluate the quality of our audio-visual captions, showing a 12.5% gain in semantic score over baselines. Using our augmented dataset, we train a Latent Diffusion Model to generate in an encodec encoding latent space. Our model is novel in the current SOTA audio generation landscape due to our generation space, text encoder, noise schedule, and attention mechanism. Together, these innovations provide competitive open-domain audio generation. The samples, models, and implementation will be at https://audiojourney.github.io.

As audio-visual systems are being deployed for safety-critical tasks such as surveillance and malicious content filtering, their robustness remains an under-studied area. Existing published work on robustness either does not scale to large-scale dataset, or does not deal with multiple modalities. This work aims to study several key questions related to multi-modal learning through the lens of robustness: 1) Are multi-modal models necessarily more robust than uni-modal models? 2) How to efficiently measure the robustness of multi-modal learning? 3) How to fuse different modalities to achieve a more robust multi-modal model? To understand the robustness of the multi-modal model in a large-scale setting, we propose a density-based metric, and a convexity metric to efficiently measure the distribution of each modality in high-dimensional latent space. Our work provides a theoretical intuition together with empirical evidence showing how multi-modal fusion affects adversarial robustness through these metrics. We further devise a mix-up strategy based on our metrics to improve the robustness of the trained model. Our experiments on AudioSet [1] and Kinetics-Sounds [2] verify our hypothesis that multi-modal models are not necessarily more robust than their uni-modal counterparts in the face of adversarial examples. We also observe our mix-up trained method could achieve as much protection as traditional adversarial training, offering a computationally cheap alternative.

After its sweeping success in vision and language tasks, pure attention-based neural architectures (e.g.DeiT) [1] are emerging to the top of audio tagging (AT) leaderboards [2], which seemingly obsoletes traditional convolutional neural networks (CNNs), feed-forward networks or recurrent networks.However, taking a closer look, there is great variability in published research, for instance, performances of models initialized with pretrained weights differ drastically from without pretraining [2], training time for a model varies from hours to weeks, and often, essences are hidden in seemingly trivial details.This urgently calls for a comprehensive study since our 1st comparison [3] is half-decade old.In this work, we perform extensive experiments on AudioSet [4] which is the largest weakly-labeled sound event dataset available, we also did analysis based on the data quality and efficiency.We compare a few state-of-the-art baselines on the AT task, and study the performance and efficiency of 2 major categories of neural architectures: CNN variants and attention-based variants.We also closely examine their optimization procedures.Our opensourced experimental results 1 provide insights to trade off between performance, efficiency, optimization process, for both practitioners and researchers. 2

This paper studies a simple extension of image-based Masked Autoencoders (MAE) to self-supervised representation learning from audio spectrograms. Following the Transformer encoder-decoder design in MAE, our Audio-MAE first encodes audio spectrogram patches with a high masking ratio, feeding only the non-masked tokens through encoder layers. The decoder then re-orders and decodes the encoded context padded with mask tokens, in order to reconstruct the input spectrogram. We find it beneficial to incorporate local window attention in the decoder, as audio spectrograms are highly correlated in local time and frequency bands. We then fine-tune the encoder with a lower masking ratio on target datasets. Empirically, Audio-MAE sets new state-of-the-art performance on six audio and speech classification tasks, outperforming other recent models that use external supervised pre-training. Our code and models is available at https://github.com/facebookresearch/AudioMAE.

As audio/visual classification models are widely deployed for sensitive tasks like content filtering at scale, it is critical to understand their robustness along with improving the accuracy. This work aims to study several key questions related to multimodal learning through the lens of adversarial noises: 1) The trade-off between early/middle/late fusion affecting its robustness and accuracy 2) How does different frequency/time domain features contribute to the robustness? 3) How does different neural modules contribute against the adversarial noise? In our experiment, we construct adversarial examples to attack state-of-the-art neural models trained on Google AudioSet.[1]1We compare how much attack potency in terms of adversarial perturbation of size using different Lpnorms we would need to "deactivate" the victim model. Using adversarial noise to ablate multimodal models, we are able to provide insights into what is the best potential fusion strategy to balance the model parameters/accuracy and robustness trade-off, and distinguish the robust features versus the non-robust features that various neural networks model tend to learn.

Automatic phonemic transcription tools are useful for low-resource language documentation. However, due to the lack of training sets, only a tiny fraction of languages have phonemic transcription tools. Fortunately, multilingual acoustic modeling provides a solution given limited audio training data. A more challenging problem is to build phonemic transcribers for languages with zero training data. The difficulty of this task is that phoneme inventories often differ between the training languages and the target language, making it infeasible to recognize unseen phonemes. In this work, we address this problem by adopting the idea of zero-shot learning. Our model is able to recognize unseen phonemes in the target language without any training data. In our model, we decompose phonemes into corresponding articulatory attributes such as vowel and consonant. Instead of predicting phonemes directly, we first predict distributions over articulatory attributes, and then compute phoneme distributions with a customized acoustic model. We evaluate our model by training it using 13 languages and testing it using 7 unseen languages. We find that it achieves 7.7% better phoneme error rate on average over a standard multilingual model.

Multilingual models can improve language processing, particularly for low resource situations, by sharing parameters across languages. Multilingual acoustic models, however, generally ignore the difference between phonemes (sounds that can support lexical contrasts in a particular language) and their corresponding phones (the sounds that are actually spoken, which are language independent). This can lead to performance degradation when combining a variety of training languages, as identically annotated phonemes can actually correspond to several different underlying phonetic realizations. In this work, we propose a joint model of both language-independent phone and language-dependent phoneme distributions. In multilingual ASR experiments over 11 languages, we find that this model improves testing performance by 2% phoneme error rate absolute in low-resource conditions. Additionally, because we are explicitly modeling language-independent phones, we can build a (nearly-)universal phone recognizer that, when combined with the PHOIBLE [1] large, manually curated database of phone inventories, can be customized into 2,000 language dependent recognizers. Experiments on two low-resourced indigenous languages, Inuktitut and Tusom, show that our recognizer achieves phone accuracy improvements of more than 17%, moving a step closer to speech recognition for all languages in the world.1

Sound event detection (SED) entails two subtasks: recognizing what types of sound events are present in an audio stream (audio tagging), and pinpointing their onset and offset times (localization). In the popular multiple instance learning (MIL) framework for SED with weak labeling, an important component is the pooling function. This paper compares five types of pooling functions both theoretically and experimentally, with special focus on their performance of localization. Although the attention pooling function is currently receiving the most attention, we find the linear softmax pooling function to perform the best among the five. Using this pooling function, we build a neural network called TALNet. It is the first system to reach state-of-the-art audio tagging performance on Audio Set, while exhibiting strong localization performance on the DCASE 2017 challenge at the same time.

Recent work has documented the susceptibility of deep learning systems to adversarial examples, but most such attacks directly manipulate the digital input to a classifier. Although a smaller line of work considers physical adversarial attacks, in all cases these involve manipulating the object of interest, e.g., putting a physical sticker on an object to misclassify it, or manufacturing an object specifically intended to be misclassified. In this work, we consider an alternative question: is it possible to fool deep classifiers, over all perceived objects of a certain type, by physically manipulating the camera itself? We show that by placing a carefully crafted and mainly-translucent sticker over the lens of a camera, one can create universal perturbations of the observed images that are inconspicuous, yet misclassify target objects as a different (targeted) class. To accomplish this, we propose an iterative procedure for both updating the attack perturbation (to make it adversarial for a given classifier), and the threat model itself (to ensure it is physically realizable). For example, we show that we can achieve physically-realizable attacks that fool ImageNet classifiers in a targeted fashion 49.6% of the time. This presents a new class of physically-realizable threat models to consider in the context of adversarially robust machine learning. Our demo video can be viewed at: https://youtu.be/wUVmL33Fx54

Voice Assistants (VAs) such as Amazon Alexa or Google Assistant rely on wake-word detection to respond to people's commands, which could potentially be vulnerable to audio adversarial examples. In this work, we target our attack on the wake-word detection system. Our goal is to jam the model with some inconspicuous background music to deactivate the VAs while our audio adversary is present. We implemented an emulated wake-word detection system of Amazon Alexa based on recent publications. We validated our models against the real Alexa in terms of wake-word detection accuracy. Then we computed our audio adversaries with consideration of expectation over transform and we implemented our audio adversary with a differentiable synthesizer. Next we verified our audio adversaries digitally on hundreds of samples of utterances collected from the real world. Our experiments show that we can effectively reduce the recognition F1 score of our emulated model from 93.4% to 11.0%. Finally, we tested our audio adversary over the air, and verified it works effectively against Alexa, reducing its F1 score from 92.5% to 11.0%. To the best of our knowledge, this is the first real-world adversarial attack against a commercial grade VA wake-word detection system. Our demo video is included in the supplementary material.

The lack of strong labels has severely limited the state-of-the-art fully supervised audio tagging systems to be scaled to larger dataset. Meanwhile, audio-visual learning models based on unlabeled videos have been successfully applied to audio tagging, but they are inevitably resource hungry and require a long time to train. In this work, we propose a light-weight, multimodal framework for environmental audio tagging. The audio branch of the framework is a convolutional and recurrent neural network (CRNN) based on multiple instance learning (MIL). It is trained with the audio tracks of a large collection of weakly labeled YouTube video excerpts; the video branch uses pretrained state-of-the-art image recognition networks and word embeddings to extract information from the video track and to map visual objects to sound events. Experiments on the audio tagging task of the DCASE 2017 challenge show that the incorporation of video information improves a strong baseline audio tagging system by 5.3% in terms of F1score. The entire system can be trained within 6 hours on a single GPU, and can be easily carried over to other audio tasks such as speech sentimental analysis.

In this paper, we introduce the concept of Eventness for audio event detection, which can, in part, be thought of as an analogue to Objectness from computer vision. The key observation behind the eventness concept is that audio events reveal themselves as 2-dimensional time-frequency patterns with specific textures and geometric structures in spectrograms. These time-frequency patterns can then be viewed analogously to objects occurring in natural images (with the exception that scaling and rotation invariance properties do not apply). With this key observation in mind, we pose the problem of detecting monophonic or polyphonic audio events as an equivalent visual object(s) detection problem under partial occlusion and clutter in spectrograms. We adapt a state-of-the-art visual object detection model to evaluate the audio event detection task on publicly available datasets. The proposed network has comparable results with a state-of-the-art baseline and is more robust on minority events. Provided large-scale datasets, we hope that our proposed conceptual model of eventness will be beneficial to the audio signal processing community towards improving performance of audio event detection.

State-of-the-art audio event detection (AED) systems rely on supervised learning using strongly labeled data. However, this dependence severely limits scalability to large-scale datasets where fine resolution annotations are too expensive to obtain. In this paper, we propose a small-footprint multiple instance learning (MIL) framework for multi-class AED using weakly annotated labels. The proposed MIL framework uses audio embeddings extracted from a pre-trained convolutional neural network as input features. We show that by using audio embeddings the MIL framework can be implemented using a simple DNN with performance comparable to recurrent neural networks. We evaluate our approach by training an audio tagging system using a subset of AudioSet, which is a large collection of weakly labeled YouTube video excerpts. Combined with a late-fusion approach, we improve the F1 score of a baseline audio tagging system by 17%. We show that audio embeddings extracted by the convolutional neural networks significantly boost the performance of all MIL models. This framework reduces the model complexity of the AED system and is suitable for applications where computational resources are limited.

Many sequence learning tasks require the localization of certain events in sequences. Because it can be expensive to obtain strong labeling that specifies the starting and ending times of the events, modern systems are often trained with weak labeling without explicit timing information. Multiple instance learning (MIL) is a popular framework for learning from weak labeling. In a common scenario of MIL, it is necessary to choose a pooling function to aggregate the predictions for the individual steps of the sequences. In this paper, we compare the and pooling functions on a speech recognition task and a sound event detection task. We find that max pooling is able to localize phonemes and sound events, while noisy-or pooling fails. We provide a theoretical explanation of the different behavior of the two pooling functions on sequence learning tasks.

Constructing a joint representation invariant across different modalities (e.g., video, language) is of significant importance in many multimedia applications. While there are a number of recent successes in developing effective image-text retrieval methods by learning joint representations, the video-text retrieval task, however, has not been explored to its fullest extent. In this paper, we study how to effectively utilize available multimodal cues from videos for the cross-modal video-text retrieval task. Based on our analysis, we propose a novel framework that simultaneously utilizes multi-modal features (different visual characteristics, audio inputs, and text) by a fusion strategy for efficient retrieval. Furthermore, we explore several loss functions in training the embedding and propose a modified pairwise ranking loss for the task. Experiments on MSVD and MSR-VTT datasets demonstrate that our method achieves significant performance gain compared to the state-of-the-art approaches.

Learning acoustic models directly from the raw waveform data with minimal processing is challenging. Current waveform-based models have generally used very few (~2) convolutional layers, which might be insufficient for building high-level discriminative features. In this work, we propose very deep convolutional neural networks (CNNs) that directly use time-domain waveforms as inputs. Our CNNs, with up to 34 weight layers, are efficient to optimize over very long sequences (e.g., vector of size 32000), necessary for processing acoustic waveforms. This is achieved through batch normalization, residual learning, and a careful design of down-sampling in the initial layers. Our networks are fully convolutional, without the use of fully connected layers and dropout, to maximize representation learning. We use a large receptive field in the first convolutional layer to mimic bandpass filters, but very small receptive fields subsequently to control the model capacity. We demonstrate the performance gains with the deeper models. Our evaluation shows that the CNN with 18 weight layers outperforms the CNN with 3 weight layers by over 15% in absolute accuracy for an environmental sound recognition task and is competitive with the performance of models using log-mel features.

Environmental sound detection is a challenging application of machine learning because of the noisy nature of the signal, and the small amount of (labeled) data that is typically available. This work thus presents a comparison of several state-of-the-art Deep Learning models on the IEEE challenge on Detection and Classification of Acoustic Scenes and Events (DCASE) 2016 challenge task and data, classifying sounds into one of fifteen common indoor and outdoor acoustic scenes, such as bus, cafe, car, city center, forest path, library, train, etc. In total, 13 hours of stereo audio recordings are available, making this one of the largest datasets available.

Occupancy estimation is an important primitive for a wide range of applications including building energy efficiency, safety, and security. In this paper, we explore the potential of using depth sensors to detect, estimate, identify, and track occupants in buildings. While depth sensors have been widely used for human detection and gesture recognition, computer vision algorithms are typically run on a powerful computer like XBOX or Intel R CoreTM i7 processor. In this work, we develop a prototype system called FORK using off-the-shelf components that performs the entire depth data processing on a cheaper and low power ARM processor in real-time. As ARM processors are extremely weak in running computer vision algorithms, FORK is designed to detect humans and track them in a very efficient way by leveraging a novel lightweight model based approach instead of traditional approaches based on histogram of oriented gradients (HOG) features. Unlike other camera based approaches, FORK is much less privacy invasive (even if the sensor is compromised). Based on a complete implementation, real-world deployment, and extensive evaluation at realistic scenarios, we observe that FORK achieves over 99% accuracy in real-time (4-9 FPS) in occupancy estimation.