Jump to Content
Aren Jansen

Aren Jansen

I am currently a Research Scientist at Google, working in the Sound Understanding Group on machine learning for speech, music and audio processing. Before joining Google, I was a Research Scientist at the Johns Hopkins University Human Language Technology Center of Excellence, an Assistant Research Professor in the John Hopkins Department of Electrical and Computer Engineering, and a faculty member of the Center for Language and Speech Processing. My research has explored a wide range of ML topics that involve unsupervised/semi-supervised representation learning, information retrieval, content-based recommendation, latent structure discovery, time series modeling and analysis, and scalable algorithms for big data applications. See my personal website or my Google scholar page for a full list of publications.
Authored Publications
Google Publications
Other Publications
Sort By
  • Title
  • Title, desc
  • Year
  • Year, desc
    MusicLM: Generating Music From Text
    Andrea Agostinelli
    Mauro Verzetti
    Antoine Caillon
    Qingqing Huang
    Neil Zeghidour
    Christian Frank
    under review (2023)
    Preview abstract We introduce MusicLM, a model generating high-fidelity music from text descriptions such as "a calming violin melody backed by a distorted guitar riff". MusicLM casts the process of conditional music generation as a hierarchical sequence-to-sequence modeling task, and it generates music at 24 kHz that remains consistent over several minutes. Our experiments show that MusicLM outperforms previous systems both in audio quality and adherence to the text description. Moreover, we demonstrate that MusicLM can be conditioned on both text and a melody in that it can transform whistled and hummed melodies according to the style described in a text caption. To support future research, we publicly release MusicCaps, a dataset composed of 5.5k music-text pairs, with rich text descriptions provided by human experts. Further links: samples, MusicCaps dataset View details
    Preview abstract Many speech applications require understanding aspects other than content, such as recognizing emotion, detecting whether the speaker is wearing a mask, or distinguishing real from synthetic speech. Generally-useful paralinguistic speech representations offer one solution to these kinds of problems. In this work, we introduce a new state-of-the-art paralinguistic speech representation based on self-supervised training of a 600M+ parameter Conformer-based architecture. Linear classifiers trained on top of our best representation outperform previous results on 7 of 8 tasks we evaluate. We perform a larger comparison than has been done previously both in terms of number of embeddings compared and number of downstream datasets evaluated on. Our analyses into the role of time demonstrate the importance of context window size for many downstream tasks. Furthermore, while the optimal representation is extracted internally in the network, we demonstrate stable high performance across several layers, allowing a single universal representation to reach near optimal performance on all tasks. View details
    Preview abstract Amyotrophic Lateral Sclerosis (ALS) disease progression is usually measured using the subjective, questionnaire-based revised ALS Functional Rating Scale (ALSFRS-R). A purely objective measure for tracking disease progression would be a powerful tool for evaluating real-world drug effectiveness, efficacy in clinical trials, as well as identifying participants for cohort studies. Here we develop a machine learning based objective measure for ALS disease progression, based on voice samples and accelerometer measurements. The ALS Therapy Development Institute (ALS-TDI) collected a unique dataset of voice and accelerometer samples from consented individuals - 584 people living with ALS over four years. Participants carried out prescribed speaking and limb-based tasks. 542 participants contributed 5814 voice recordings, and 350 contributed 13009 accelerometer samples, while simultaneously measuring ALSFRS-R. Using the data from 475 participants, we trained machine learning (ML) models, correlating voice with bulbar-related FRS scores and accelerometer with limb related scores. On the test set (n=109 participants) the voice models achieved an AUC of 0.86 (95% CI, 0.847-0.884) , whereas the accelerometer models achieved a median AUC of 0.73 . We used the models and self-reported ALSFRS-R scores to evaluate the real-world effects of edaravone, a drug recently approved for use in ALS, on 54 test participants. In the test cohort, the digital data input into the ML models produced objective measures of progression rates over the duration of the study that were consistent with self-reported scores. This demonstrates the value of these tools for assessing both disease progression and potentially drug effects. In this instance, outcomes from edaravone treatment, both self-reported and digital-ML, resulted in highly variable outcomes from person to person. View details
    MuLan: A Joint Embedding of Music Audio and Natural Language
    Qingqing Huang
    Ravi Ganti
    Judith Yue Li
    Proceedings of the the 23rd International Society for Music Information Retrieval Conference (ISMIR) (2022) (to appear)
    Preview abstract Music tagging and content-based retrieval systems have traditionally been constructed using pre-defined ontologies covering a rigid set of music attributes or text queries. This paper presents MuLan: a first attempt at a new generation of acoustic models that link music audio directly to unconstrained natural language music descriptions. MuLan takes the form of a two-tower, joint audio-text embedding model trained using 44 million music recordings (370K hours) and weakly-associated, free-form text annotations. Through its compatibility with a wide range of music genres and text styles (including conventional music tags), the resulting audio-text representation subsumes existing ontologies while graduating to true zero-shot functionalities. We demonstrate the versatility of the MuLan embeddings with a range of experiments including transfer learning, zero-shot music tagging, language understanding in the music domain, and cross-modal retrieval applications. View details
    Preview abstract Recent progress in deep learning has enabled many advances in sound separation and visual scene understanding. However, extracting sound sources which are apparent in natural videos remains an open problem. In this work, we present AudioScope, a novel audio-visual sound separation framework that can be trained without supervision to isolate on-screen sound sources from real in-the-wild videos. Prior audio-visual separation work assumed artificial limitations on the domain of sound classes (e.g., to speech or music), constrained the number of sources, and required strong sound separation or visual segmentation labels. AudioScope overcomes these limitations, operating on an open domain of sounds, with variable numbers of sources, and without labels or prior visual segmentation. The training procedure for AudioScope uses mixture invariant training (MixIT) to separate synthetic mixtures of mixtures (MoMs) into individual sources, where noisy labels for mixtures are provided by an unsupervised audio-visual coincidence model. Using the noisy labels, along with attention between video and audio features, AudioScope learns to identify audio-visual similarity and to suppress off-screen sounds. We demonstrate the effectiveness of our approach using a dataset of video clips extracted from open-domain YFCC100m video data. This dataset contains a wide diversity of sound classes recorded in unconstrained conditions, making the application of previous methods unsuitable. For evaluation and semi-supervised experiments, we collected human labels for presence of on-screen and off-screen sounds on a small subset of clips. View details
    Preview abstract Supervised neural network training has led to significant progress on single-channel sound separation. This approach relies on ground truth isolated sources, which precludes scaling to widely available mixture data and limits progress on open-domain tasks. The recent mixture invariant training (MixIT) method enables training on in-the-wild data; however, it suffers from two outstanding problems. First, it produces models which tend to over-separate, producing more output sources than are present in the input. Second, the exponential computational complexity of the MixIT loss limits the number of feasible output sources. In this paper we address both issues. To combat over-separation we introduce new losses: sparsity losses that favor fewer output sources and a covariance loss that discourages correlated outputs. We also experiment with a semantic classification loss by predicting weak class labels for each mixture. To handle larger numbers of sources, we introduce an efficient approximation using a fast least-squares solution, projected onto the MixIT constraint set. Our experiments show that the proposed losses curtail over-separation and improve overall performance. The best performance is achieved using larger numbers of output sources, enabled by our efficient MixIT loss, combined with sparsity losses to prevent over-separation. On the FUSS test set, we achieve over 13 dB in multi-source SI-SNR improvement, while boosting single-source reconstruction SI-SNR by over 17 dB. View details
    Preview abstract Humans perceive the world by concurrently processing and fusing high-dimensional inputs from multiple modalities such as vision and audio. Machine perception models, in stark contrast, are typically modality-specific and optimised for unimodal benchmarks. A common approach for building multimodal models is to simply combine multiple of these modality-specific architectures using late-stage fusion of final representations or predictions (\textit{`late-fusion'}). Instead, we propose a new architecture that learns to model both unimodal and cross-modal information at earlier stages, without imposing any modality specific priors. We investigate two pathways for the exchange of cross-modal information, \textit{vertical attention} (by restricting crossmodal fusion to certain layers) and \textit{horizontal attention}, via the use of `fusion bottleneck' tokens, that encourage the model to extract and exchange relevant information between modalities in an efficient manner. We conduct thorough ablation studies, and achieve state-of-the-art results on multiple audio-visual classification benchmarks including Audioset, Epic-Kitchens and VGGSound. All code and models will be released. View details
    A Convolutional Neural Network for Automated Detection of Humpback Whale Song in a Diverse, Long-Term Passive Acoustic Dataset
    Ann N. Allen
    Matt Harvey
    Karlina P. Merkens
    Carrie C. Wall
    Erin M. Oleson
    Frontiers in Marine Science, vol. 8 (2021), pp. 165
    Preview abstract Passive acoustic monitoring is a well-established tool for researching the occurrence, movements, and ecology of a wide variety of marine mammal species. Advances in hardware and data collection have exponentially increased the volumes of passive acoustic data collected, such that discoveries are now limited by the time required to analyze rather than collect the data. In order to address this limitation, we trained a deep convolutional neural network (CNN) to identify humpback whale song in over 187,000 h of acoustic data collected at 13 different monitoring sites in the North Pacific over a 14-year period. The model successfully detected 75 s audio segments containing humpback song with an average precision of 0.97 and average area under the receiver operating characteristic curve (AUC-ROC) of 0.992. The model output was used to analyze spatial and temporal patterns of humpback song, corroborating known seasonal patterns in the Hawaiian and Mariana Islands, including occurrence at remote monitoring sites beyond well-studied aggregations, as well as novel discovery of humpback whale song at Kingman Reef, at 5∘ North latitude. This study demonstrates the ability of a CNN trained on a small dataset to generalize well to a highly variable signal type across a diverse range of recording and noise conditions. We demonstrate the utility of active learning approaches for creating high-quality models in specialized domains where annotations are rare. These results validate the feasibility of applying deep learning models to identify highly variable signals across broad spatial and temporal scales, enabling new discoveries through combining large datasets with cutting edge tools. View details
    Preview abstract To reveal the importance of temporal precision in ground truth audio event labels, we collected precise (∼0.1 sec resolution) “strong” labels for a portion of the AudioSet dataset. We devised a temporally strong evaluation set (including explicit negatives of varying difficulty) and a small strong-labeled training subset of 67k clips (compared to the original dataset’s 1.8M clips labeled at 10 sec resolution). We show that fine-tuning with a mix of weak- and strongly-labeled data can substantially improve classifier performance, even when evaluated using only the original weak labels. For a ResNet50 architecture, d' on the strong evaluation data including explicit negatives improves from 1.13 to 1.41. The new labels are available as an update to AudioSet. View details
    Preview abstract Real-world sound scenes consist of time-varying collections of sound sources, each generating characteristic sound events that are mixed together in audio recordings. The association of these constituent sound events with their mixture and each other is semantically-constrained: the sound scene contains the union of source classes and not all classes naturally co-occur. With this motivation, this paper explores the use of unsupervised automatic sound separation to decompose unlabeled sound scenes into multiple semantically-linked views for use in self-supervised contrastive learning. We find that learning to associate input mixtures with their automatically separated outputs yields stronger representations than past approaches that use the mixtures alone. Further, we discover that optimal source separation is not required for successful contrastive learning by demonstrating that a range of separation system convergence states all lead to useful and often complementary example transformations. Our best system incorporates these unsupervised separation models into a single augmentation front-end and jointly optimizes similarity maximization and coincidence prediction objectives across the views. The result is an unsupervised audio representation that rivals state-of-the-art alternatives on the established shallow AudioSet classification benchmark. View details
    Preview abstract The ultimate goal of transfer learning is to enable learning with a small amount of data, by using a strong embedding. While significant progress has been made in the visual and language domains, the speech domain does not have such a universal method. This paper presents a new representation of speech signals based on an unsupervised triplet-loss objective, which outperforms both existing state of the art and other representations on a number of transfer learning tasks in the non-semantic speech domain. The embedding is learned on a publicly available dataset, and it is tested on a variety of low-resource downstream tasks, including personalization tasks and medical domain. The model will be publicly released. View details
    Preview abstract Deep learning approaches have recently achieved impressive performance on both audio source separation and sound classification. Most audio source separation approaches focus only on separating sources belonging to a restricted domain of source classes, such as speech and music. However, recent work has demonstrated the possibility of "universal sound separation", which aims to separate acoustic sources from an open domain, regardless of their class. In this paper, we utilize the semantic information learned by sound classifier networks trained on a vast amount of diverse sounds to improve universal sound separation. In particular, we show that semantic embeddings extracted from a sound classifier can be used to condition a separation network, providing it with useful additional information. This approach is especially useful in an iterative setup, where source estimates from an initial separation stage and their corresponding classifier-derived embeddings are fed to a second separation network. By performing a thorough hyperparameter search consisting of over a thousand experiments, we find that classifier embeddings from oracle clean sources provide nearly one dB of SNR gain, and our best iterative models achieve a significant fraction of this oracle performance, establishing a new state-of-the-art for universal sound separation. View details
    Preview abstract Recent progress in deep learning has enabled many advances in sound separation and visual scene understanding. However, extracting sound sources which are apparent in natural videos remains an open problem. In this work, we present AudioScope, a novel audio-visual sound separation framework that can be trained without supervision to isolate on-screen sound sources from real in-the-wild videos. Prior audio-visual separation work assumed artificial limitations on the domain of sound classes (e.g., to speech or music), constrained the number of sources, and required strong sound separation or visual segmentation labels. AudioScope overcomes these limitations, operating on an open domain of sounds, with variable numbers of sources, and without labels or prior visual segmentation. The training procedure for AudioScope uses mixture invariant training (MixIT) to separate synthetic mixtures of mixtures (MoMs) into individual sources, where noisy labels for mixtures are provided by an unsupervised audio-visual coincidence model. Using the noisy labels, along with attention between video and audio features, AudioScope learns to identify audio-visual similarity and to suppress off-screen sounds. We demonstrate the effectiveness of our approach using a dataset of video clips extracted from open-domain YFCC100m video data. This dataset contains a wide diversity of sound classes recorded in unconstrained conditions, making the application of previous methods unsuitable. For evaluation and semi-supervised experiments, we collected human labels for presence of on-screen and off-screen sounds on a small subset of clips. View details
    Preview abstract Humans do not acquire perceptual abilities like we train machines. While machine learning algorithms typically operate on large collections of randomly-chosen, explicitly-labeled examples, human acquisition relies far greater on multimodal unsupervised learning (as infants) and active learning (as children). With this motivation, we present a learning framework for sound representation and recognition that combines (i) a self-supervised objective based on a general notion of unimodal and cross-modal coincidence, (ii) a novel clustering objective that reflects our need to impose categorical structure on our experiences, and (iii) a cluster-based active learning procedure that solicits targeted weak supervision to consolidate hypothesized categories into relevant semantic classes. By jointly training a single sound embedding/clustering/classification network according to these criteria, we achieve a new state-of-the-art unsupervised audio representation and demonstrate up to 20-fold reduction in labels required to reach a desired classification performance. View details
    Preview abstract We explore content-based representation learning strategies tailored for large-scale, uncurated music collections that afford only weak supervision through unstructured natural language metadata and co-listen statistics. At the core is a hybrid training scheme that uses classification and metric learning losses to incorporate both metadata-derived text labels and aggregate co-listen supervisory signals into a single convolutional model. The resulting joint text and audio content embedding defines a similarity metric and supports prediction of semantic text labels using a vocabulary of unprecedented granularity, which we refine using a novel word-sense disambiguation procedure. As input to simple classifier architectures, our representation achieves state-of-the-art performance on two music tagging benchmarks. View details
    Preview abstract Even in the absence of any explicit semantic annotation, vast collections of audio recordings provide valuable information for learning the categorical structure of sounds. We consider several class-agnostic semantic constraints that apply to unlabeled nonspeech audio: (i) noise and translations in time do not change the underlying sound category, (ii) a mixture of two sound events inherits the categories of the constituents, and (iii) the categories of events in close temporal proximity are likely to be the same or related. Without labels to ground them, these constraints are incompatible with classification loss functions. However, they may still be leveraged to identify geometric inequalities needed for triplet loss-based training of convolutional neural networks. The result is low-dimensional embeddings of the input spectrograms that recover 41% and 84% of the performance of their fully-supervised counterparts when applied to downstream query-by-example sound retrieval and sound event classification tasks, respectively. Moreover, in limited-supervision settings, our unsupervised embeddings double the state-of-the-art classification performance. View details
    A Segmental Framework for Fully-Unsupervised Large-Vocabulary Speech Recognition
    Herman Kamper
    Sharon Goldwater
    Computer Speech and Language (2017) (to appear)
    Preview abstract Zero-resource speech technology is a growing research area that aims to develop methods for speech processing in the absence of transcriptions, lexicons, or language modelling text. Early systems focused on identifying isolated recurring terms in a corpus, while more recent full-coverage systems attempt to completely segment and cluster the audio into word-like units—effectively performing unsupervised speech recognition. To our knowledge, this article presents the first such system evaluated on large vocabulary multi-speaker data. The system uses a Bayesian modelling framework with segmental word representations: each word segment is represented as a fixed-dimensional acoustic embedding obtained by mapping the sequence of feature frames to a single embedding vector. We compare our system on English and Xitsonga datasets to state-of-the-art baselines, using a variety of measures including word error rate (obtained by mapping the unsupervised output to ground truth transcriptions). We show that by imposing a consistent top-down segmentation while also using bottom-up knowledge from detected syllable boundaries, both single-speaker and multi-speaker versions of our system outperform a purely bottom-up single-speaker syllable-based approach. We also show that the discovered clusters can be made less speaker- and gender-specific by using an unsupervised autoencoder-like feature extractor to learn better frame-level features (prior to embedding). Our system’s discovered clusters are still less pure than those of two multi-speaker term discovery systems, but provide far greater coverage. View details
    Towards Learning Semantic Audio Representations from Unlabeled Data
    Ratheet Pandya
    Jiayang Liu
    NIPS Workshop on Machine Learning for Audio Signal Processing (ML4Audio) (2017) (to appear)
    Preview abstract Our goal is to learn semantically structured audio representations without relying on categorically labeled data. We consider several class-agnostic semantic constraints that are inherent to non-speech audio: (i) sound categories are invariant to additive noise and translations in time, (ii) mixtures of two sound events inherit the categories of the constituents, and (iii) the categories of events in close temporal proximity in a single recording are likely to be the same or related. We apply these invariants in the service of sampling training data for triplet-loss embedding models using a large unlabeled dataset of YouTube soundtracks. The resulting low-dimensional representations provide both greatly improved query-by-example retrieval performance and reduced labeled data and model complexity requirements for supervised sound classification. View details
    Preview abstract Convolutional Neural Networks (CNNs) have proven very effective in image classification and have shown promise for audio classification. We apply various CNN architectures to audio and investigate their ability to classify videos with a very large scale data set of 70M training videos (5.24 million hours) with 30,871 labels. We examine fully connected Deep Neural Networks (DNNs), AlexNet [1], VGG [2], Inception [3], and ResNet [4]. We explore the effects of training with different sized subsets of the 70M training videos. Additionally we report the effect of training over different subsets of the 30,871 labels. While our dataset contains video-level labels, we are also interested in Acoustic Event Detection (AED) and train a classifier on embeddings learned from the video-level task on AudioSet [5]. We find that derivatives of image classification networks do well on our audio classification task, that increasing the number of labels we train on provides some improved performance over subsets of labels, that performance of models improves as we increase training set size, and that a model using embeddings learned from the video-level task do much better than a baseline on the AudioSet classification task. View details
    Preview abstract Audio event recognition, the human-like ability to identify and relate sounds from audio, is a nascent problem in machine perception. Comparable problems such as object detection in images have reaped enormous benefits from comprehensive datasets -- principally ImageNet. This paper describes the creation of Audio Set, a large-scale dataset of manually-annotated audio events that endeavors to bridge the gap in data availability between image and audio research. Using a carefully structured hierarchical ontology of 635 audio classes guided by the literature and manual curation, we collect data from human labelers to probe the presence of specific audio classes in 10 second segments of YouTube videos. Segments are proposed for labeling using searches based on metadata, context (e.g., links), and content analysis. The result is a dataset of unprecedented breadth and size that will, we hope, substantially stimulate the development of high-performance audio event recognizers. View details
    Large-Scale Audio Event Discovery in One Million YouTube Videos
    Jort F. Gemmeke
    Xiaofeng Liu
    Wade Lawrence
    Dylan Freedman
    Proceedings of ICASSP (2017) (to appear)
    Preview abstract Internet videos provide a virtually boundless source of audio with a conspicuous lack of localized annotations, presenting an ideal setting for unsupervised methods. With this motivation, we perform an unprecedented exploration into the large-scale discovery of recurring audio events in a diverse corpus of one million YouTube videos (45K hours of audio). Our approach is to apply a streaming, nonparametric clustering algorithm to both spectral features and out-of-domain neural audio embeddings, and use a small portion of manually annotated audio events to quantitatively estimate the intrinsic clustering performance. In addition to providing a useful mechanism for unsupervised active learning, we demonstrate the effectiveness of the discovered audio event clusters in two downstream applications. The first is weakly-supervised learning, where we exploit the association of video-level metadata and cluster occurrences to temporally localize audio events. The second is informative activity detection, an unsupervised method for semantic saliency based on the corpus statistics of the discovered event clusters. View details
    Unsupervised Word Segmentation and Lexicon Discovery Using Acoustic Word Embeddings
    Herman Kamper
    Sharon Goldwater
    IEEE Transactions on Audio, Speech, and Language Processing (2016)
    Preview abstract In settings where only unlabelled speech data is available, speech technology needs to be developed without transcriptions, pronunciation dictionaries, or language modelling text. A similar problem is faced when modelling infant language acquisition. In these cases, categorical linguistic structure needs to be discovered directly from speech audio. We present a novel unsupervised Bayesian model that segments unlabelled speech and clusters the segments into hypothesized word groupings. The result is a complete unsupervised tokenization of the input speech in terms of discovered word types. In our approach, a potential word segment (of arbitrary length) is embedded in a fixed-dimensional acoustic vector space. The model, implemented as a Gibbs sampler, then builds a whole-word acoustic model in this space while jointly performing segmentation. We report word error rates in a small-vocabulary connected digit recognition task by mapping the unsupervised decoded output to ground truth transcriptions. The model achieves around 20% error rate, outperforming a previous HMM-based system by about 10% absolute. Moreover, in contrast to the baseline, our model does not require a pre-specified vocabulary size. View details
    No Results Found