Music datasets play a crucial role in advancing research in machine learning for music. However, existing music datasets suffer from limited size, accessibility, and lack of audio resources. To address these shortcomings, we present DISCO-10M, a novel and extensive music dataset that surpasses the largest previously available music dataset by an order of magnitude. To ensure high-quality data, we implement a multi-stage filtering process. This process incorporates similarities based on textual descriptions and audio embeddings. Moreover, we provide precomputed CLAP embeddings alongside DISCO-10M, facilitating direct application on various downstream tasks. These embeddings enable efficient exploration of machine learning applications on the provided data. With DISCO-10M, we aim to democratize and facilitate new research to help advance the development of novel machine learning models for music.
相關內容
The mixture proportions of pretraining data domains (e.g., Wikipedia, books, web text) greatly affect language model (LM) performance. In this paper, we propose Domain Reweighting with Minimax Optimization (DoReMi), which first trains a small proxy model using group distributionally robust optimization (Group DRO) over domains to produce domain weights (mixture proportions) without knowledge of downstream tasks. We then resample a dataset with these domain weights and train a larger, full-sized model. In our experiments, we use DoReMi on a 280M-parameter proxy model to set the domain weights for training an 8B-parameter model (30x larger) more efficiently. On The Pile, DoReMi improves perplexity across all domains, even when it downweights a domain. DoReMi improves average few-shot downstream accuracy by 6.5% points over a baseline model trained using The Pile's default domain weights and reaches the baseline accuracy with 2.6x fewer training steps. On the GLaM dataset, DoReMi, which has no knowledge of downstream tasks, even matches the performance of using domain weights tuned on downstream tasks.
The generative AI revolution has recently expanded to videos. Nevertheless, current state-of-the-art video models are still lagging behind image models in terms of visual quality and user control over the generated content. In this work, we present a framework that harnesses the power of a text-to-image diffusion model for the task of text-driven video editing. Specifically, given a source video and a target text-prompt, our method generates a high-quality video that adheres to the target text, while preserving the spatial layout and motion of the input video. Our method is based on a key observation that consistency in the edited video can be obtained by enforcing consistency in the diffusion feature space. We achieve this by explicitly propagating diffusion features based on inter-frame correspondences, readily available in the model. Thus, our framework does not require any training or fine-tuning, and can work in conjunction with any off-the-shelf text-to-image editing method. We demonstrate state-of-the-art editing results on a variety of real-world videos. Webpage: //diffusion-tokenflow.github.io/
We present a novel single-stage framework, Neural Photon Field (NePF), to address the ill-posed inverse rendering from multi-view images. Contrary to previous methods that recover the geometry, material, and illumination in multiple stages and extract the properties from various multi-layer perceptrons across different neural fields, we question such complexities and introduce our method - a single-stage framework that uniformly recovers all properties. NePF achieves this unification by fully utilizing the physical implication behind the weight function of neural implicit surfaces and the view-dependent radiance. Moreover, we introduce an innovative coordinate-based illumination model for rapid volume physically-based rendering. To regularize this illumination, we implement the subsurface scattering model for diffuse estimation. We evaluate our method on both real and synthetic datasets. The results demonstrate the superiority of our approach in recovering high-fidelity geometry and visual-plausible material attributes.
We present Radarize, a self-contained SLAM pipeline for indoor environments that uses only a low-cost commodity single-chip mmWave radar. Our radar-native approach leverages phenomena unique to radio frequencies, such as doppler shift-based odometry, to improve performance. We evaluate our method on a large-scale dataset of 146 trajectories spanning 4 campus buildings, totaling approximately 4680m of travel distance. Our results show that our method outperforms state-of-the-art radar-based approaches by approximately 5x in terms of odometry and 8x in terms of end-to-end SLAM, as measured by absolute trajectory error (ATE), without the need additional sensors such as IMUs or wheel odometry.
Audio diffusion models can synthesize a wide variety of sounds. Existing models often operate on the latent domain with cascaded phase recovery modules to reconstruct waveform. This poses challenges when generating high-fidelity audio. In this paper, we propose EDMSound, a diffusion-based generative model in spectrogram domain under the framework of elucidated diffusion models (EDM). Combining with efficient deterministic sampler, we achieved similar Fr\'echet audio distance (FAD) score as top-ranked baseline with only 10 steps and reached state-of-the-art performance with 50 steps on the DCASE2023 foley sound generation benchmark. We also revealed a potential concern regarding diffusion based audio generation models that they tend to generate samples with high perceptual similarity to the data from training data. Project page: //agentcooper2002.github.io/EDMSound/
The research in the field of music is rapidly growing, and this trend emphasizes the need for comprehensive data. Though researchers have made an effort to contribute their own datasets, many data collections lack the requisite inclusivity for comprehensive study because they are frequently focused on particular components of music or other specific topics. We have endeavored to address data scarcity by employing an instrument-based approach to provide a complete corpus related to the Persian piano. Our piano corpus includes relevant labels for Persian music mode (Dastgah) and comprehensive metadata, allowing for utilization in various popular research areas. The features extracted from 2022 Persian piano pieces in The Persian Piano Corpus (PPC) have been collected and made available to researchers, aiming for a more thorough understanding of Persian music and the role of the piano in it in subsequent steps.
This study addresses the deficiency in conventional music recommendation systems by focusing on the vital role of emotions in shaping users music choices. These systems often disregard the emotional context, relying predominantly on past listening behavior and failing to consider the dynamic and evolving nature of users emotional preferences. This gap leads to several limitations. Users may receive recommendations that do not match their current mood, which diminishes the quality of their music experience. Furthermore, without accounting for emotions, the systems might overlook undiscovered or lesser-known songs that have a profound emotional impact on users. To combat these limitations, this research introduces an AI model that incorporates emotional context into the song recommendation process. By accurately detecting users real-time emotions, the model can generate personalized song recommendations that align with the users emotional state. This approach aims to enhance the user experience by offering music that resonates with their current mood, elicits the desired emotions, and creates a more immersive and meaningful listening experience. By considering emotional context in the song recommendation process, the proposed model offers an opportunity for a more personalized and emotionally resonant musical journey.
We present CoDEx, a set of knowledge graph completion datasets extracted from Wikidata and Wikipedia that improve upon existing knowledge graph completion benchmarks in scope and level of difficulty. In terms of scope, CoDEx comprises three knowledge graphs varying in size and structure, multilingual descriptions of entities and relations, and tens of thousands of hard negative triples that are plausible but verified to be false. To characterize CoDEx, we contribute thorough empirical analyses and benchmarking experiments. First, we analyze each CoDEx dataset in terms of logical relation patterns. Next, we report baseline link prediction and triple classification results on CoDEx for five extensively tuned embedding models. Finally, we differentiate CoDEx from the popular FB15K-237 knowledge graph completion dataset by showing that CoDEx covers more diverse and interpretable content, and is a more difficult link prediction benchmark. Data, code, and pretrained models are available at //bit.ly/2EPbrJs.
We present MMKG, a collection of three knowledge graphs that contain both numerical features and (links to) images for all entities as well as entity alignments between pairs of KGs. Therefore, multi-relational link prediction and entity matching communities can benefit from this resource. We believe this data set has the potential to facilitate the development of novel multi-modal learning approaches for knowledge graphs.We validate the utility ofMMKG in the sameAs link prediction task with an extensive set of experiments. These experiments show that the task at hand benefits from learning of multiple feature types.
We construct targeted audio adversarial examples on automatic speech recognition. Given any audio waveform, we can produce another that is over 99.9% similar, but transcribes as any phrase we choose (at a rate of up to 50 characters per second). We apply our iterative optimization-based attack to Mozilla's implementation DeepSpeech end-to-end, and show it has a 100% success rate. The feasibility of this attack introduce a new domain to study adversarial examples.
小貼士
登錄享
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191