Barry-John Theobald

Bhavika Devnani, Skyler Seto, Zakaria Aldeneh et al.

Humans can picture a sound scene given an imprecise natural language description. For example, it is easy to imagine an acoustic environment given a phrase like "the lion roar came from right behind me!". For a machine to have the same degree of comprehension, the machine must know what a lion is (semantic attribute), what the concept of "behind" is (spatial attribute) and how these pieces of linguistic information align with the semantic and spatial attributes of the sound (what a roar sounds like when its coming from behind). State-of-the-art audio foundation models which learn to map between audio scenes and natural textual descriptions, are trained on non-spatial audio and text pairs, and hence lack spatial awareness. In contrast, sound event localization and detection models are limited to recognizing sounds from a fixed number of classes, and they localize the source to absolute position (e.g., 0.2m) rather than a position described using natural language (e.g., "next to me"). To address these gaps, we present ELSA a spatially aware-audio and text embedding model trained using multimodal contrastive learning. ELSA supports non-spatial audio, spatial audio, and open vocabulary text captions describing both the spatial and semantic components of sound. To train ELSA: (a) we spatially augment the audio and captions of three open-source audio datasets totaling 4,738 hours of audio, and (b) we design an encoder to capture the semantics of non-spatial audio, and the semantics and spatial attributes of spatial audio using contrastive learning. ELSA is competitive with state-of-the-art for both semantic retrieval and 3D source localization. In particular, ELSA achieves +2.8% mean audio-to-text and text-to-audio R@1 above the baseline, and outperforms by -11.6° mean-absolute-error in 3D source localization over the baseline.

Skyler Seto, Barry-John Theobald, Federico Danieli et al.

Fully-test-time adaptation (F-TTA) can mitigate performance loss due to distribution shifts between train and test data (1) without access to the training data, and (2) without knowledge of the model training procedure. In online F-TTA, a pre-trained model is adapted using a stream of test samples by minimizing a self-supervised objective, such as entropy minimization. However, models adapted with online using entropy minimization, are unstable especially in single sample settings, leading to degenerate solutions, and limiting the adoption of TTA inference strategies. Prior works identify noisy, or unreliable, samples as a cause of failure in online F-TTA. One solution is to ignore these samples, which can lead to bias in the update procedure, slow adaptation, and poor generalization. In this work, we present a general framework for improving robustness of F-TTA to these noisy samples, inspired by self-paced learning and robust loss functions. Our proposed approach, Robust Entropy Adaptive Loss Minimization (REALM), achieves better adaptation accuracy than previous approaches throughout the adaptation process on corruptions of CIFAR-10 and ImageNet-1K, demonstrating its effectiveness.

Zakaria Aldeneh, Masha Fedzechkina, Skyler Seto et al. · apple-ml

Generating realistic lip motion from audio to simulate speech production is critical for driving natural character animation. Previous research has shown that traditional metrics used to optimize and assess models for generating lip motion from speech are not a good indicator of subjective opinion of animation quality. Devising metrics that align with subjective opinion first requires understanding what impacts human perception of quality. In this work, we focus on the degree of articulation and run a series of experiments to study how articulation strength impacts human perception of lip motion accompanying speech. Specifically, we study how increasing under-articulated (dampened) and over-articulated (exaggerated) lip motion affects human perception of quality. We examine the impact of articulation strength on human perception when considering only lip motion, where viewers are presented with talking faces represented by landmarks, and in the context of embodied characters, where viewers are presented with photo-realistic videos. Our results show that viewers prefer over-articulated lip motion consistently more than under-articulated lip motion and that this preference generalizes across different speakers and embodiments.

Akshay Mehra, Skyler Seto, Navdeep Jaitly et al.

Multi-Exit models (MEMs) use an early-exit strategy to improve the accuracy and efficiency of deep neural networks (DNNs) by allowing samples to exit the network before the last layer. However, the effectiveness of MEMs in the presence of distribution shifts remains largely unexplored. Our work examines how distribution shifts generated by common image corruptions affect the accuracy/efficiency of MEMs. We find that under common corruptions, early-exiting at the first correct exit reduces the inference cost and provides a significant boost in accuracy ( 10%) over exiting at the last layer. However, with realistic early-exit strategies, which do not assume knowledge about the correct exits, MEMs still reduce inference cost but provide a marginal improvement in accuracy (1%) compared to exiting at the last layer. Moreover, the presence of distribution shift widens the gap between an MEM's maximum classification accuracy and realistic early-exit strategies by 5% on average compared with the gap on in-distribution data. Our empirical analysis shows that the lack of calibration due to a distribution shift increases the susceptibility of such early-exit strategies to exit early and increases misclassification rates. Furthermore, the lack of calibration increases the inconsistency in the predictions of the model across exits, leading to both inefficient inference and more misclassifications compared with evaluation on in-distribution data. Finally, we propose two metrics, underthinking and overthinking, that quantify the different behavior of practical early-exit strategy under distribution shifts, and provide insights into improving the practical utility of MEMs.

Yong Lin, Skyler Seto, Maartje ter Hoeve et al.

Reinforcement Learning from Human Feedback (RLHF) is an effective approach for aligning language models to human preferences. Central to RLHF is learning a reward function for scoring human preferences. Two main approaches for learning a reward model are 1) training an EXplicit Reward Model (EXRM) as in RLHF, and 2) using an implicit reward learned from preference data through methods such as Direct Preference Optimization (DPO). Prior work has shown that the implicit reward model of DPO (denoted as DPORM) can approximate an EXRM in the limit. DPORM's effectiveness directly implies the optimality of the learned policy, and also has practical implication for LLM alignment methods including iterative DPO. However, it is unclear how well DPORM empirically matches the performance of EXRM. This work studies the accuracy at distinguishing preferred and rejected answers for both DPORM and EXRM. Our findings indicate that even though DPORM fits the training dataset comparably, it generalizes less effectively than EXRM, especially when the validation datasets contain distribution shifts. Across five out-of-distribution settings, DPORM has a mean drop in accuracy of 3% and a maximum drop of 7%. These findings highlight that DPORM has limited generalization ability and substantiates the integration of an explicit reward model in iterative DPO approaches.

Miguel Sarabia, Elena Menyaylenko, Alessandro Toso et al.

We present Spatial LibriSpeech, a spatial audio dataset with over 650 hours of 19-channel audio, first-order ambisonics, and optional distractor noise. Spatial LibriSpeech is designed for machine learning model training, and it includes labels for source position, speaking direction, room acoustics and geometry. Spatial LibriSpeech is generated by augmenting LibriSpeech samples with 200k+ simulated acoustic conditions across 8k+ synthetic rooms. To demonstrate the utility of our dataset, we train models on four spatial audio tasks, resulting in a median absolute error of 6.60° on 3D source localization, 0.43m on distance, 90.66ms on T30, and 2.74dB on DRR estimation. We show that the same models generalize well to widely-used evaluation datasets, e.g., obtaining a median absolute error of 12.43° on 3D source localization on TUT Sound Events 2018, and 157.32ms on T30 estimation on ACE Challenge.

Katherine Metcalf, Miguel Sarabia, Barry-John Theobald

Preference-based reinforcement learning (RL) algorithms help avoid the pitfalls of hand-crafted reward functions by distilling them from human preference feedback, but they remain impractical due to the burdensome number of labels required from the human, even for relatively simple tasks. In this work, we demonstrate that encoding environment dynamics in the reward function (REED) dramatically reduces the number of preference labels required in state-of-the-art preference-based RL frameworks. We hypothesize that REED-based methods better partition the state-action space and facilitate generalization to state-action pairs not included in the preference dataset. REED iterates between encoding environment dynamics in a state-action representation via a self-supervised temporal consistency task, and bootstrapping the preference-based reward function from the state-action representation. Whereas prior approaches train only on the preference-labelled trajectory pairs, REED exposes the state-action representation to all transitions experienced during policy training. We explore the benefits of REED within the PrefPPO [1] and PEBBLE [2] preference learning frameworks and demonstrate improvements across experimental conditions to both the speed of policy learning and the final policy performance. For example, on quadruped-walk and walker-walk with 50 preference labels, REED-based reward functions recover 83% and 66% of ground truth reward policy performance and without REED only 38\% and 21\% are recovered. For some domains, REED-based reward functions result in policies that outperform policies trained on the ground truth reward.

Zakaria Aldeneh, Skyler Seto, Maureen de Seyssel et al.

Modern ASR systems are typically trained on large-scale pseudo-labeled, in-the-wild data spanning multiple domains. While such heterogeneous data benefit generalist models designed for broad deployment, they pose challenges for specialist models targeting specific domains: specialist models lack the capacity to learn from all available data, and one must pay closer attention to addressing the mismatch between training and test conditions. In this work, we study targeted data selection as a strategy to address these challenges, selecting relevant subsets from 100k hours of in-the-wild training data to optimize performance on target domains. We represent speech samples using embeddings that capture complementary characteristic--speaker attributes, phonetic content, and semantic meaning--and analyze how relevance and diversity along these axes when performing data selection affect downstream ASR performance. Our experiments with CTC-based Conformer models show that training on a strategically selected 5% subset can exceed the performance of models trained on the full dataset by up to 36.8% relative WER reduction on target domains.

Jee-weon Jung, Wangyou Zhang, Jiatong Shi et al.

This paper introduces ESPnet-SPK, a toolkit designed with several objectives for training speaker embedding extractors. First, we provide an open-source platform for researchers in the speaker recognition community to effortlessly build models. We provide several models, ranging from x-vector to recent SKA-TDNN. Through the modularized architecture design, variants can be developed easily. We also aspire to bridge developed models with other domains, facilitating the broad research community to effortlessly incorporate state-of-the-art embedding extractors. Pre-trained embedding extractors can be accessed in an off-the-shelf manner and we demonstrate the toolkit's versatility by showcasing its integration with two tasks. Another goal is to integrate with diverse self-supervised learning features. We release a reproducible recipe that achieves an equal error rate of 0.39% on the Vox1-O evaluation protocol using WavLM-Large with ECAPA-TDNN.

Nico Lingg, Miguel Sarabia, Luca Zappella et al.

Human skeleton point clouds are commonly used to automatically classify and predict the behaviour of others. In this paper, we use a contrastive self-supervised learning method, SimCLR, to learn representations that capture the semantics of skeleton point clouds. This work focuses on systematically evaluating the effects that different algorithmic decisions (including augmentations, dataset partitioning and backbone architecture) have on the learned skeleton representations. To pre-train the representations, we normalise six existing datasets to obtain more than 40 million skeleton frames. We evaluate the quality of the learned representations with three downstream tasks: skeleton reconstruction, motion prediction, and activity classification. Our results demonstrate the importance of 1) combining spatial and temporal augmentations, 2) including additional datasets for encoder training, and 3) and using a graph neural network as an encoder.

Katherine Metcalf, Miguel Sarabia, Natalie Mackraz et al.

Preference-based reinforcement learning (PbRL) aligns a robot behavior with human preferences via a reward function learned from binary feedback over agent behaviors. We show that dynamics-aware reward functions improve the sample efficiency of PbRL by an order of magnitude. In our experiments we iterate between: (1) learning a dynamics-aware state-action representation (z^{sa}) via a self-supervised temporal consistency task, and (2) bootstrapping the preference-based reward function from (z^{sa}), which results in faster policy learning and better final policy performance. For example, on quadruped-walk, walker-walk, and cheetah-run, with 50 preference labels we achieve the same performance as existing approaches with 500 preference labels, and we recover 83\% and 66\% of ground truth reward policy performance versus only 38\% and 21\%. The performance gains demonstrate the benefits of explicitly learning a dynamics-aware reward model. Repo: \texttt{https://github.com/apple/ml-reed}.

Yinong Oliver Wang, Nivedha Sivakumar, Falaah Arif Khan et al.

The recent rapid adoption of large language models (LLMs) highlights the critical need for benchmarking their fairness. Conventional fairness metrics, which focus on discrete accuracy-based evaluations (i.e., prediction correctness), fail to capture the implicit impact of model uncertainty (e.g., higher model confidence about one group over another despite similar accuracy). To address this limitation, we propose an uncertainty-aware fairness metric, UCerF, to enable a fine-grained evaluation of model fairness that is more reflective of the internal bias in model decisions compared to conventional fairness measures. Furthermore, observing data size, diversity, and clarity issues in current datasets, we introduce a new gender-occupation fairness evaluation dataset with 31,756 samples for co-reference resolution, offering a more diverse and suitable dataset for evaluating modern LLMs. We establish a benchmark, using our metric and dataset, and apply it to evaluate the behavior of ten open-source LLMs. For example, Mistral-7B exhibits suboptimal fairness due to high confidence in incorrect predictions, a detail overlooked by Equalized Odds but captured by UCerF. Overall, our proposed LLM benchmark, which evaluates fairness with uncertainty awareness, paves the way for developing more transparent and accountable AI systems.

Lucas Monteiro Paes, Natalie Mackraz, Barry-John Theobald et al.

Language model (LM) alignment improves model outputs to reflect human preferences while preserving the capabilities of the base model. The most common alignment approaches are (i) reinforcement learning, which maximizes the expected reward under a KL-divergence constraint, and (ii) best-of-$N$ alignment, which selects the highest-reward output among $N$ independent samples. Despite their widespread use, the fundamental limits of reward improvement under a KL budget remain poorly understood. We characterize the information-theoretic limits of KL-regularized alignment by deriving the maximum achievable expected reward gain for a fixed KL-divergence budget. Our first result provides a closed-form expression for the optimal reward improvement, governed by a Jeffreys divergence term rather than the $\sqrt{\texttt{KL}}$ used in prior analyses. We further reformulate this expression as a covariance under the base model, yielding a practical estimator that predicts achievable alignment gains from base model samples alone. We extend our analysis to the proxy reward setting, showing that the gap between ideal and proxy alignment (reward hacking) grows with the magnitude of reward error and when the KL penalty factor decreases. We then prove that reward ensembling mitigates reward hacking, providing a theoretical justification for this technique used in practice. Empirically, we compute the KL-reward Pareto frontier for two tasks for LMs, safety and summarization, and show that best-of-$N$ closely approaches the theoretical limit, while PPO and GRPO remain substantially suboptimal. Our theoretical results shed light on several empirically observed phenomena in the alignment literature and suggest that algorithmic improvements are needed to achieve optimal alignment without high inference costs.

Anirudh Sundar, Sinead Williamson, Katherine Metcalf et al. · apple-ml

Aligned representations across languages is a desired property in multilingual large language models (mLLMs), as alignment can improve performance in cross-lingual tasks. Typically alignment requires fine-tuning a model, which is computationally expensive, and sizable language data, which often may not be available. A data-efficient alternative to fine-tuning is model interventions -- a method for manipulating model activations to steer generation into the desired direction. We analyze the effect of a popular intervention (finding experts) on the alignment of cross-lingual representations in mLLMs. We identify the neurons to manipulate for a given language and introspect the embedding space of mLLMs pre- and post-manipulation. We show that modifying the mLLM's activations changes its embedding space such that cross-lingual alignment is enhanced. Further, we show that the changes to the embedding space translate into improved downstream performance on retrieval tasks, with up to 2x improvements in top-1 accuracy on cross-lingual retrieval.

Stéphane Aroca-Ouellette, Natalie Mackraz, Barry-John Theobald et al.

Accommodating human preferences is essential for creating aligned LLM agents that deliver personalized and effective interactions. Recent work has shown the potential for LLMs acting as writing agents to infer a description of user preferences. Agent alignment then comes from conditioning on the inferred preference description. However, existing methods often produce generic preference descriptions that fail to capture the unique and individualized nature of human preferences. This paper introduces PROSE, a method designed to enhance the precision of preference descriptions inferred from user writing samples. PROSE incorporates two key elements: (1) iterative refinement of inferred preferences, and (2) verification of inferred preferences across multiple user writing samples. We evaluate PROSE with several LLMs (i.e., Qwen2.5 7B and 72B Instruct, GPT-mini, and GPT-4o) on a summarization and an email writing task. We find that PROSE more accurately infers nuanced human preferences, improving the quality of the writing agent's generations over CIPHER (a state-of-the-art method for inferring preferences) by 33\%. Lastly, we demonstrate that ICL and PROSE are complementary methods, and combining them provides up to a 9\% improvement over ICL alone.

Natalie Mackraz, Nivedha Sivakumar, Samira Khorshidi et al.

Large language models (LLMs) are increasingly being adapted to achieve task-specificity for deployment in real-world decision systems. Several previous works have investigated the bias transfer hypothesis (BTH) by studying the effect of the fine-tuning adaptation strategy on model fairness to find that fairness in pre-trained masked language models have limited effect on the fairness of models when adapted using fine-tuning. In this work, we expand the study of BTH to causal models under prompt adaptations, as prompting is an accessible, and compute-efficient way to deploy models in real-world systems. In contrast to previous works, we establish that intrinsic biases in pre-trained Mistral, Falcon and Llama models are strongly correlated (rho >= 0.94) with biases when the same models are zero- and few-shot prompted, using a pronoun co-reference resolution task. Further, we find that bias transfer remains strongly correlated even when LLMs are specifically prompted to exhibit fair or biased behavior (rho >= 0.92), and few-shot length and stereotypical composition are varied (rho >= 0.97). Our findings highlight the importance of ensuring fairness in pre-trained LLMs, especially when they are later used to perform downstream tasks via prompt adaptation.

Nivedha Sivakumar, Natalie Mackraz, Samira Khorshidi et al.

A dangerous assumption that can be made from prior work on the bias transfer hypothesis (BTH) is that biases do not transfer from pre-trained large language models (LLMs) to adapted models. We invalidate this assumption by studying the BTH in causal models under prompt adaptations, as prompting is an extremely popular and accessible adaptation strategy used in real-world applications. In contrast to prior work, we find that biases can transfer through prompting and that popular prompt-based mitigation methods do not consistently prevent biases from transferring. Specifically, the correlation between intrinsic biases and those after prompt adaptation remain moderate to strong across demographics and tasks -- for example, gender (rho >= 0.94) in co-reference resolution, and age (rho >= 0.98) and religion (rho >= 0.69) in question answering. Further, we find that biases remain strongly correlated when varying few-shot composition parameters, such as sample size, stereotypical content, occupational distribution and representational balance (rho >= 0.90). We evaluate several prompt-based debiasing strategies and find that different approaches have distinct strengths, but none consistently reduce bias transfer across models, tasks or demographics. These results demonstrate that correcting bias, and potentially improving reasoning ability, in intrinsic models may prevent propagation of biases to downstream tasks.

Masha Fedzechkina, Eleonora Gualdoni, Sinead Williamson et al. · apple-ml

Activation steering methods in large language models (LLMs) have emerged as an effective way to perform targeted updates to enhance generated language without requiring large amounts of adaptation data. We ask whether the features discovered by activation steering methods are interpretable. We identify neurons responsible for specific concepts (e.g., ``cat'') using the ``finding experts'' method from research on activation steering and show that the ExpertLens, i.e., inspection of these neurons provides insights about model representation. We find that ExpertLens representations are stable across models and datasets and closely align with human representations inferred from behavioral data, matching inter-human alignment levels. ExpertLens significantly outperforms the alignment captured by word/sentence embeddings. By reconstructing human concept organization through ExpertLens, we show that it enables a granular view of LLM concept representation. Our findings suggest that ExpertLens is a flexible and lightweight approach for capturing and analyzing model representations.

Lucas Monteiro Paes, Nivedha Sivakumar, Yinong Oliver Wang et al.

Generative models are often deployed to make decisions on behalf of users, such as vision-language models (VLMs) identifying which person in a room is a doctor to help visually impaired individuals. Yet, VLM decisions are influenced by the perceived demographic attributes of people in the input, which can lead to biased outcomes like failing to identify women as doctors. Moreover, when reducing bias leads to performance loss, users may have varying needs for balancing bias mitigation with overall model capabilities, highlighting the demand for methods that enable controllable bias reduction during inference. Activation steering is a popular approach for inference-time controllability that has shown potential in inducing safer behavior in large language models (LLMs). However, we observe that current steering methods struggle to correct biases, where equiprobable outcomes across demographic groups are required. To address this, we propose Direct Steering Optimization (DSO) which uses reinforcement learning to find linear transformations for steering activations, tailored to mitigate bias while maintaining control over model performance. We demonstrate that DSO achieves state-of-the-art trade-off between fairness and capabilities on both VLMs and LLMs, while offering practitioners inference-time control over the trade-off. Overall, our work highlights the benefit of designing steering strategies that are directly optimized to control model behavior, providing more effective bias intervention than methods that rely on pre-defined heuristics for controllability.

Falaah Arif Khan, Nivedha Sivakumar, Yinong Oliver Wang et al.

Large language models (LLMs) have achieved impressive performance, leading to their widespread adoption as decision-support tools in resource-constrained contexts like hiring and admissions. There is, however, scientific consensus that AI systems can reflect and exacerbate societal biases, raising concerns about identity-based harm when used in critical social contexts. Prior work has laid a solid foundation for assessing bias in LLMs by evaluating demographic disparities in different language reasoning tasks. In this work, we extend single-axis fairness evaluations to examine intersectional bias, recognizing that when multiple axes of discrimination intersect, they create distinct patterns of disadvantage. We create a new benchmark called WinoIdentity by augmenting the WinoBias dataset with 25 demographic markers across 10 attributes, including age, nationality, and race, intersected with binary gender, yielding 245,700 prompts to evaluate 50 distinct bias patterns. Focusing on harms of omission due to underrepresentation, we investigate bias through the lens of uncertainty and propose a group (un)fairness metric called Coreference Confidence Disparity which measures whether models are more or less confident for some intersectional identities than others. We evaluate five recently published LLMs and find confidence disparities as high as 40% along various demographic attributes including body type, sexual orientation and socio-economic status, with models being most uncertain about doubly-disadvantaged identities in anti-stereotypical settings. Surprisingly, coreference confidence decreases even for hegemonic or privileged markers, indicating that the recent impressive performance of LLMs is more likely due to memorization than logical reasoning. Notably, these are two independent failures in value alignment and validity that can compound to cause social harm.

Krishna Patel, Nivedha Sivakumar, Barry-John Theobald et al.

We investigate fairness dynamics during Large Language Model (LLM) training to enable the diagnoses of biases and mitigations through training interventions like early stopping; we find that biases can emerge suddenly and do not always follow common performance metrics. We introduce two new metrics to evaluate fairness dynamics holistically during model pre-training: Average Rank and Jensen-Shannon Divergence by Parts. These metrics provide insights into the Pythia models' progression of biases in gender prediction of occupations on the WinoBias dataset. By monitoring these dynamics, we find that (1) Pythia-6.9b is biased towards men; it becomes more performant and confident predicting "male" than "female" during training, (2) via early-stopping, Pythia-6.9b can exchange 1.7% accuracy on LAMBADA for a 92.5% increase in fairness, and (3) larger models can exhibit more bias; Pythia-6.9b makes more assumptions about gender than Pythia-160m, even when a subject's gender is not specified.

Chen Huang, Skyler Seto, Hadi Pouransari et al. · utoronto

Vision foundation models pre-trained on massive data encode rich representations of real-world concepts, which can be adapted to downstream tasks by fine-tuning. However, fine-tuning foundation models on one task often leads to the issue of concept forgetting on other tasks. Recent methods of robust fine-tuning aim to mitigate forgetting of prior knowledge without affecting the fine-tuning performance. Knowledge is often preserved by matching the original and fine-tuned model weights or feature pairs. However, such point-wise matching can be too strong, without explicit awareness of the feature neighborhood structures that encode rich knowledge as well. We propose a novel regularization method Proxy-FDA that explicitly preserves the structural knowledge in feature space. Proxy-FDA performs Feature Distribution Alignment (using nearest neighbor graphs) between the pre-trained and fine-tuned feature spaces, and the alignment is further improved by informative proxies that are generated dynamically to increase data diversity. Experiments show that Proxy-FDA significantly reduces concept forgetting during fine-tuning, and we find a strong correlation between forgetting and a distributional distance metric (in comparison to L2 distance). We further demonstrate Proxy-FDA's benefits in various fine-tuning settings (end-to-end, few-shot and continual tuning) and across different tasks like image classification, captioning and VQA.

Andrew Silva, Katherine Metcalf, Nicholas Apostoloff et al.

Federated learning enables the deployment of machine learning to problems for which centralized data collection is impractical. Adding differential privacy guarantees bounds on privacy while data are contributed to a global model. Adding personalization to federated learning introduces new challenges as we must account for preferences of individual users, where a data sample could have conflicting labels because one sub-population of users might view an input positively, but other sub-populations view the same input negatively. We present FedEmbed, a new approach to private federated learning for personalizing a global model that uses (1) sub-populations of similar users, and (2) personal embeddings. We demonstrate that current approaches to federated learning are inadequate for handling data with conflicting labels, and we show that FedEmbed achieves up to 45% improvement over baseline approaches to personalized private federated learning.

Andrew Silva, Barry-John Theobald, Nicholas Apostoloff

Automatic speech recognition (ASR) is widely used in consumer electronics. ASR greatly improves the utility and accessibility of technology, but usually the output is only word sequences without punctuation. This can result in ambiguity in inferring user-intent. We first present a transformer-based approach for punctuation prediction that achieves 8% improvement on the IWSLT 2012 TED Task, beating the previous state of the art [1]. We next describe our multimodal model that learns from both text and audio, which achieves 8% improvement over the text-only algorithm on an internal dataset for which we have both the audio and transcriptions. Finally, we present an approach to learning a model using contextual dropout that allows us to handle variable amounts of future context at test time.

Nataniel Ruiz, Barry-John Theobald, Anurag Ranjan et al.

To detect bias in face recognition networks, it can be useful to probe a network under test using samples in which only specific attributes vary in some controlled way. However, capturing a sufficiently large dataset with specific control over the attributes of interest is difficult. In this work, we describe a simulator that applies specific head pose and facial expression adjustments to images of previously unseen people. The simulator first fits a 3D morphable model to a provided image, applies the desired head pose and facial expression controls, then renders the model into an image. Next, a conditional Generative Adversarial Network (GAN) conditioned on the original image and the rendered morphable model is used to produce the image of the original person with the new facial expression and head pose. We call this conditional GAN -- MorphGAN. Images generated using MorphGAN conserve the identity of the person in the original image, and the provided control over head pose and facial expression allows test sets to be created to identify robustness issues of a facial recognition deep network with respect to pose and expression. Images generated by MorphGAN can also serve as data augmentation when training data are scarce. We show that by augmenting small datasets of faces with new poses and expressions improves the recognition performance by up to 9% depending on the augmentation and data scarcity.

Ahmed Hussen Abdelaziz, Barry-John Theobald, Paul Dixon et al.

We describe our novel deep learning approach for driving animated faces using both acoustic and visual information. In particular, speech-related facial movements are generated using audiovisual information, and non-speech facial movements are generated using only visual information. To ensure that our model exploits both modalities during training, batches are generated that contain audio-only, video-only, and audiovisual input features. The probability of dropping a modality allows control over the degree to which the model exploits audio and visual information during training. Our trained model runs in real-time on resource limited hardware (e.g.\ a smart phone), it is user agnostic, and it is not dependent on a potentially error-prone transcription of the speech. We use subjective testing to demonstrate: 1) the improvement of audiovisual-driven animation over the equivalent video-only approach, and 2) the improvement in the animation of speech-related facial movements after introducing modality dropout. Before introducing dropout, viewers prefer audiovisual-driven animation in 51% of the test sequences compared with only 18% for video-driven. After introducing dropout viewer preference for audiovisual-driven animation increases to 74%, but decreases to 8% for video-only.

Zakaria Aldeneh, Anushree Prasanna Kumar, Barry-John Theobald et al.

We present an introspection of an audiovisual speech enhancement model. In particular, we focus on interpreting how a neural audiovisual speech enhancement model uses visual cues to improve the quality of the target speech signal. We show that visual cues provide not only high-level information about speech activity, i.e., speech/silence, but also fine-grained visual information about the place of articulation. One byproduct of this finding is that the learned visual embeddings can be used as features for other visual speech applications. We demonstrate the effectiveness of the learned visual embeddings for classifying visemes (the visual analogy to phonemes). Our results provide insight into important aspects of audiovisual speech enhancement and demonstrate how such models can be used for self-supervision tasks for visual speech applications.

Ahmed Hussen Abdelaziz, Barry-John Theobald, Justin Binder et al.

Speech-driven visual speech synthesis involves mapping features extracted from acoustic speech to the corresponding lip animation controls for a face model. This mapping can take many forms, but a powerful approach is to use deep neural networks (DNNs). However, a limitation is the lack of synchronized audio, video, and depth data required to reliably train the DNNs, especially for speaker-independent models. In this paper, we investigate adapting an automatic speech recognition (ASR) acoustic model (AM) for the visual speech synthesis problem. We train the AM on ten thousand hours of audio-only data. The AM is then adapted to the visual speech synthesis domain using ninety hours of synchronized audio-visual speech. Using a subjective assessment test, we compared the performance of the AM-initialized DNN to one with a random initialization. The results show that viewers significantly prefer animations generated from the AM-initialized DNN than the ones generated using the randomly initialized model. We conclude that visual speech synthesis can significantly benefit from the powerful representation of speech in the ASR acoustic models.

Katherine Metcalf, Barry-John Theobald, Garrett Weinberg et al.

We describe experiments towards building a conversational digital assistant that considers the preferred conversational style of the user. In particular, these experiments are designed to measure whether users prefer and trust an assistant whose conversational style matches their own. To this end we conducted a user study where subjects interacted with a digital assistant that responded in a way that either matched their conversational style, or did not. Using self-reported personality attributes and subjects' feedback on the interactions, we built models that can reliably predict a user's preferred conversational style.

Katherine Metcalf, Barry-John Theobald, Nicholas Apostoloff

We propose a method for modeling and learning turn-taking behaviors for accessing a shared resource. We model the individual behavior for each agent in an interaction and then use a multi-agent fusion model to generate a summary over the expected actions of the group to render the model independent of the number of agents. The individual behavior models are weighted finite state transducers (WFSTs) with weights dynamically updated during interactions, and the multi-agent fusion model is a logistic regression classifier. We test our models in a multi-agent tower-building environment, where a Q-learning agent learns to interact with rule-based agents. Our approach accurately models the underlying behavior patterns of the rule-based agents with accuracy ranging between 0.63 and 1.0 depending on the stochasticity of the other agent behaviors. In addition we show using KL-divergence that the model accurately captures the distribution of next actions when interacting with both a single agent (KL-divergence < 0.1) and with multiple agents (KL-divergence < 0.37). Finally, we demonstrate that our behavior model can be used by a Q-learning agent to take turns in an interactive turn-taking environment.

Helen L. Bear, Richard W. Harvey, Barry-John Theobald et al.

A critical assumption of all current visual speech recognition systems is that there are visual speech units called visemes which can be mapped to units of acoustic speech, the phonemes. Despite there being a number of published maps it is infrequent to see the effectiveness of these tested, particularly on visual-only lip-reading (many works use audio-visual speech). Here we examine 120 mappings and consider if any are stable across talkers. We show a method for devising maps based on phoneme confusions from an automated lip-reading system, and we present new mappings that show improvements for individual talkers.

Helen L. Bear, Gari Owen, Richard Harvey et al.

In the quest for greater computer lip-reading performance there are a number of tacit assumptions which are either present in the datasets (high resolution for example) or in the methods (recognition of spoken visual units called visemes for example). Here we review these and other assumptions and show the surprising result that computer lip-reading is not heavily constrained by video resolution, pose, lighting and other practical factors. However, the working assumption that visemes, which are the visual equivalent of phonemes, are the best unit for recognition does need further examination. We conclude that visemes, which were defined over a century ago, are unlikely to be optimal for a modern computer lip-reading system.

Helen L. Bear, Richard Harvey, Barry-John Theobald et al.

Visual-only speech recognition is dependent upon a number of factors that can be difficult to control, such as: lighting; identity; motion; emotion and expression. But some factors, such as video resolution are controllable, so it is surprising that there is not yet a systematic study of the effect of resolution on lip-reading. Here we use a new data set, the Rosetta Raven data, to train and test recognizers so we can measure the affect of video resolution on recognition accuracy. We conclude that, contrary to common practice, resolution need not be that great for automatic lip-reading. However it is highly unlikely that automatic lip-reading can work reliably when the distance between the bottom of the lower lip and the top of the upper lip is less than four pixels at rest.