Chanwoo Kim

Ian Covert, Chanwoo Kim, Su-In Lee

Transformers have become a default architecture in computer vision, but understanding what drives their predictions remains a challenging problem. Current explanation approaches rely on attention values or input gradients, but these provide a limited view of a model's dependencies. Shapley values offer a theoretically sound alternative, but their computational cost makes them impractical for large, high-dimensional models. In this work, we aim to make Shapley values practical for vision transformers (ViTs). To do so, we first leverage an attention masking approach to evaluate ViTs with partial information, and we then develop a procedure to generate Shapley value explanations via a separate, learned explainer model. Our experiments compare Shapley values to many baseline methods (e.g., attention rollout, GradCAM, LRP), and we find that our approach provides more accurate explanations than existing methods for ViTs.

Chris Lin, Hugh Chen, Chanwoo Kim et al.

Despite the widespread use of unsupervised models, very few methods are designed to explain them. Most explanation methods explain a scalar model output. However, unsupervised models output representation vectors, the elements of which are not good candidates to explain because they lack semantic meaning. To bridge this gap, recent works defined a scalar explanation output: a dot product-based similarity in the representation space to the sample being explained (i.e., an explicand). Although this enabled explanations of unsupervised models, the interpretation of this approach can still be opaque because similarity to the explicand's representation may not be meaningful to humans. To address this, we propose contrastive corpus similarity, a novel and semantically meaningful scalar explanation output based on a reference corpus and a contrasting foil set of samples. We demonstrate that contrastive corpus similarity is compatible with many post-hoc feature attribution methods to generate COntrastive COrpus Attributions (COCOA) and quantitatively verify that features important to the corpus are identified. We showcase the utility of COCOA in two ways: (i) we draw insights by explaining augmentations of the same image in a contrastive learning setting (SimCLR); and (ii) we perform zero-shot object localization by explaining the similarity of image representations to jointly learned text representations (CLIP).

Kisang Park, Chanwoo Kim, Kyungjae Lee et al.

Generating collision-free and smooth motions remains a central challenge in robotic manipulation, particularly in cluttered environments and narrow passages where feasible regions are highly constrained and fragmented. We propose a trajectory optimization framework that performs geometry-aware updates directly in function space using natural functional gradients. The method optimizes a Gaussian-smoothed surrogate objective that regularizes the optimization landscape through smooth trajectory perturbations while preserving trajectory-level structure. Because the updates are defined intrinsically in function space, trajectory regularity can be controlled independently of a particular time discretization. We derive a practical Monte-Carlo estimator of the natural functional gradient that requires only black-box trajectory evaluations, making the method applicable when analytic gradients are unavailable or unreliable due to collision checking and contact-rich simulation. Experiments on constrained robotic manipulation tasks demonstrate that the proposed method improves trajectory feasibility and produces smoother motions than representative planning and trajectory optimization baselines in environments with narrow geometric clearances. Additional results, videos, and implementation details are available at the project page: https://kisangpark.github.io/natural-functional-gradient/

Eunseop Yoon, Hee Suk Yoon, Dhananjaya Gowda et al.

Text-to-Text Transfer Transformer (T5) has recently been considered for the Grapheme-to-Phoneme (G2P) transduction. As a follow-up, a tokenizer-free byte-level model based on T5 referred to as ByT5, recently gave promising results on word-level G2P conversion by representing each input character with its corresponding UTF-8 encoding. Although it is generally understood that sentence-level or paragraph-level G2P can improve usability in real-world applications as it is better suited to perform on heteronyms and linking sounds between words, we find that using ByT5 for these scenarios is nontrivial. Since ByT5 operates on the character level, it requires longer decoding steps, which deteriorates the performance due to the exposure bias commonly observed in auto-regressive generation models. This paper shows that the performance of sentence-level and paragraph-level G2P can be improved by mitigating such exposure bias using our proposed loss-based sampling method.

Hoseong Cho, Donguk Kim, Chanwoo Kim et al.

This report describes our 1st place solution to ECCV 2022 challenge on Human Body, Hands, and Activities (HBHA) from Egocentric and Multi-view Cameras (hand pose estimation). In this challenge, we aim to estimate global 3D hand poses from the input image where two hands and an object are interacting on the egocentric viewpoint. Our proposed method performs end-to-end multi-hand pose estimation via transformer architecture. In particular, our method robustly estimates hand poses in a scenario where two hands interact. Additionally, we propose an algorithm that considers hand scales to robustly estimate the absolute depth. The proposed algorithm works well even when the hand sizes are various for each person. Our method attains 14.4 mm (left) and 15.9 mm (right) errors for each hand in the test set.

Jihwan Lee, Joun Yeop Lee, Heejin Choi et al.

Intonations play an important role in delivering the intention of a speaker. However, current end-to-end TTS systems often fail to model proper intonations. To alleviate this problem, we propose a novel, intuitive method to synthesize speech in different intonations using predefined intonation templates. Prior to TTS model training, speech data are grouped into intonation templates in an unsupervised manner. Two proposed modules are added to the end-to-end TTS framework: an intonation predictor and an intonation encoder. The intonation predictor recommends a suitable intonation template to the given text. The intonation encoder, attached to the text encoder output, synthesizes speech abiding the requested intonation template. Main contributions of our paper are: (a) an easy-to-use intonation control system covering a wide range of users; (b) better performance in wrapping speech in a requested intonation with improved objective and subjective evaluation; and (c) incorporating a pre-trained language model for intonation modelling. Audio samples are available at https://srtts.github.io/IntoTTS.

Chanwoo Kim, Sathish Indurti, Jinhwan Park et al.

This paper proposes a new regularization algorithm referred to as macro-block dropout. The overfitting issue has been a difficult problem in training large neural network models. The dropout technique has proven to be simple yet very effective for regularization by preventing complex co-adaptations during training. In our work, we define a macro-block that contains a large number of units from the input to a Recurrent Neural Network (RNN). Rather than applying dropout to each unit, we apply random dropout to each macro-block. This algorithm has the effect of applying different drop out rates for each layer even if we keep a constant average dropout rate, which has better regularization effects. In our experiments using Recurrent Neural Network-Transducer (RNN-T), this algorithm shows relatively 4.30 % and 6.13 % Word Error Rates (WERs) improvement over the conventional dropout on LibriSpeech test-clean and test-other. With an Attention-based Encoder-Decoder (AED) model, this algorithm shows relatively 4.36 % and 5.85 % WERs improvement over the conventional dropout on the same test sets.

Joshua Tian Jin Tee, Kang Zhang, Hee Suk Yoon et al.

Diffusion models have recently emerged as a potent tool in generative modeling. However, their inherent iterative nature often results in sluggish image generation due to the requirement for multiple model evaluations. Recent progress has unveiled the intrinsic link between diffusion models and Probability Flow Ordinary Differential Equations (ODEs), thus enabling us to conceptualize diffusion models as ODE systems. Simultaneously, Physics Informed Neural Networks (PINNs) have substantiated their effectiveness in solving intricate differential equations through implicit modeling of their solutions. Building upon these foundational insights, we introduce Physics Informed Distillation (PID), which employs a student model to represent the solution of the ODE system corresponding to the teacher diffusion model, akin to the principles employed in PINNs. Through experiments on CIFAR 10 and ImageNet 64x64, we observe that PID achieves performance comparable to recent distillation methods. Notably, it demonstrates predictable trends concerning method-specific hyperparameters and eliminates the need for synthetic dataset generation during the distillation process. Both of which contribute to its easy-to-use nature as a distillation approach for Diffusion Models. Our code and pre-trained checkpoint are publicly available at: https://github.com/pantheon5100/pid_diffusion.git.

Yongjoon Lee, Chanwoo Kim

Speech Super-Resolution (SSR) is a task of enhancing low-resolution speech signals by restoring missing high-frequency components. Conventional approaches typically reconstruct log-mel features, followed by a vocoder that generates high-resolution speech in the waveform domain. However, as mel features lack phase information, this can result in performance degradation during the reconstruction phase. Motivated by recent advances with Selective State Spaces Models (SSMs), we propose a method, referred to as Wave-U-Mamba that directly performs SSR in time domain. In our comparative study, including models such as WSRGlow, NU-Wave 2, and AudioSR, Wave-U-Mamba demonstrates superior performance, achieving the lowest Log-Spectral Distance (LSD) across various low-resolution sampling rates, ranging from 8 to 24 kHz. Additionally, subjective human evaluations, scored using Mean Opinion Score (MOS) reveal that our method produces SSR with natural and human-like quality. Furthermore, Wave-U-Mamba achieves these results while generating high-resolution speech over nine times faster than baseline models on a single A100 GPU, with parameter sizes less than 2\% of those in the baseline models.

Gio Paik, Yongbeom Kim, Soungmin Lee et al.

Despite advances in multilingual automatic speech recognition (ASR), code-switching (CS), the mixing of languages within an utterance common in daily speech, remains a severely underexplored challenge. In this paper, we introduce HiKE: the Hierarchical Korean-English code-switching benchmark, the first globally accessible evaluation framework for Korean-English CS, aiming to provide a means for the precise evaluation of multilingual ASR models and to foster research in the field. The proposed framework not only consists of high-quality, natural CS data across various topics, but also provides meticulous loanword labels and a hierarchical CS-level labeling scheme (word, phrase, and sentence) that together enable a systematic evaluation of a model's ability to handle each distinct level of code-switching. Through evaluations of diverse multilingual ASR models and fine-tuning experiments, this paper demonstrates that although most multilingual ASR models initially exhibit inadequate CS-ASR performance, this capability can be enabled through fine-tuning with synthetic CS data. HiKE is available at https://github.com/ThetaOne-AI/HiKE

Chanwoo Kim, Jinkyu Sung, Yebonn Han et al.

Graph convolutional networks have recently gained prominence in collaborative filtering (CF) for recommendations. However, we identify potential bottlenecks in two foundational components. First, the embedding layer leads to a latent space with limited capacity, overlooking locally observed but potentially valuable preference patterns. Also, the widely-used neighborhood aggregation is limited in its ability to leverage diverse preference patterns in a fine-grained manner. Building on spectral graph theory, we reveal that these limitations stem from graph filtering with a cut-off in the frequency spectrum and a restricted linear form. To address these issues, we introduce ChebyCF, a CF framework based on graph spectral filtering. Instead of a learned embedding, it takes a user's raw interaction history to utilize the full spectrum of signals contained in it. Also, it adopts Chebyshev interpolation to effectively approximate a flexible non-linear graph filter, and further enhances it by using an additional ideal pass filter and degree-based normalization. Through extensive experiments, we verify that ChebyCF overcomes the aforementioned bottlenecks and achieves state-of-the-art performance across multiple benchmarks and reasonably fast inference. Our code is available at https://github.com/chanwoo0806/ChebyCF.

Wei Chu, Yuanzhe Dong, Ke Tan et al.

OleSpeech-IV dataset is a large-scale multispeaker and multilingual conversational speech dataset with diverse topics. The audio content comes from publicly-available English podcasts, talk shows, teleconferences, and other conversations. Speaker names, turns, and transcripts are human-sourced and refined by a proprietary pipeline, while additional information such as timestamps and confidence scores is derived from the pipeline. The IV denotes its position as Tier IV in the Olewave dataset series. In addition, we have open-sourced a subset, OleSpeech-IV-2025-EN-AR-100, for non-commercial research use.

Chanwoo Kim, Ehsan Variani, Arun Narayanan et al.

In this paper, we describe how to efficiently implement an acoustic room simulator to generate large-scale simulated data for training deep neural networks. Even though Google Room Simulator in [1] was shown to be quite effective in reducing the Word Error Rates (WERs) for far-field applications by generating simulated far-field training sets, it requires a very large number of Fast Fourier Transforms (FFTs) of large size. Room Simulator in [1] used approximately 80 percent of Central Processing Unit (CPU) usage in our CPU + Graphics Processing Unit (GPU) training architecture [2]. In this work, we implement an efficient OverLap Addition (OLA) based filtering using the open-source FFTW3 library. Further, we investigate the effects of the Room Impulse Response (RIR) lengths. Experimentally, we conclude that we can cut the tail portions of RIRs whose power is less than 20 dB below the maximum power without sacrificing the speech recognition accuracy. However, we observe that cutting RIR tail more than this threshold harms the speech recognition accuracy for rerecorded test sets. Using these approaches, we were able to reduce CPU usage for the room simulator portion down to 9.69 percent in CPU/GPU training architecture. Profiling result shows that we obtain 22.4 times speed-up on a single machine and 37.3 times speed up on Google's distributed training infrastructure.

Mingyu Lu, Soham Gadgil, Chris Lin et al.

As Text-to-Image (T2I) diffusion models are increasingly used in real-world creative workflows, a principled framework for valuing contributors who provide a collection of data is essential for fair compensation and sustainable data marketplaces. While the Shapley value offers a theoretically grounded approach to attribution, it faces a dual computational bottleneck: (i) the prohibitive cost of exhaustive model retraining for each sampled subset of players (i.e., data contributors) and (ii) the combinatorial number of subsets needed to estimate marginal contributions due to contributor interactions. To this end, we propose SurrogateSHAP, a retraining-free framework that approximates the expensive retraining game through inference from a pretrained model. To further improve efficiency, we employ a gradient-boosted tree to approximate the utility function and derive Shapley values analytically from the tree-based model. We evaluate SurrogateSHAP across three diverse attribution tasks: (i) image quality for DDPM-CFG on CIFAR-20, (ii) aesthetics for Stable Diffusion on Post-Impressionist artworks, and (iii) product diversity for FLUX.1 on Fashion-Product data. Across settings, SurrogateSHAP outperforms prior methods while substantially reducing computational overhead, consistently identifying influential contributors across multiple utility metrics. Finally, we demonstrate that SurrogateSHAP effectively localizes data sources responsible for spurious correlations in clinical images, providing a scalable path toward auditing safety-critical generative models.

Ian Covert, Chanwoo Kim, Su-In Lee et al.

Many tasks in explainable machine learning, such as data valuation and feature attribution, perform expensive computation for each data point and are intractable for large datasets. These methods require efficient approximations, and although amortizing the process by learning a network to directly predict the desired output is a promising solution, training such models with exact labels is often infeasible. We therefore explore training amortized models with noisy labels, and we find that this is inexpensive and surprisingly effective. Through theoretical analysis of the label noise and experiments with various models and datasets, we show that this approach tolerates high noise levels and significantly accelerates several feature attribution and data valuation methods, often yielding an order of magnitude speedup over existing approaches.

Jihwan Yoon, Taemoon Jeong, Jeongeun Park et al.

Designing robot morphologies and kinematics has traditionally relied on human intuition, with little systematic foundation. Motion-design co-optimization offers a promising path toward automation, but two major challenges remain: (i) the vast, unstructured design space and (ii) the difficulty of constructing task-specific loss functions. We propose a new paradigm that minimizes human involvement by (i) learning the design search space from existing mechanical designs, rather than hand-crafting it, and (ii) defining the loss directly from human motion data via motion retargeting and Procrustes analysis. Using screw-theory-based joint axis representation and isometric manifold learning, we construct a compact, geometry-preserving latent space of humanoid upper body designs in which optimization is tractable. We then solve design optimization in this latent space using gradient-free optimization. Our approach establishes a principled framework for data-driven robot design and demonstrates that leveraging existing designs and human motion can effectively guide the automated discovery of novel robot design.

Junsu Kim, Sumin Hong, Chanwoo Kim et al.

Class incremental learning aims to solve a problem that arises when continuously adding unseen class instances to an existing model This approach has been extensively studied in the context of image classification; however its applicability to object detection is not well established yet. Existing frameworks using replay methods mainly collect replay data without considering the model being trained and tend to rely on randomness or the number of labels of each sample. Also, despite the effectiveness of the replay, it was not yet optimized for the object detection task. In this paper, we introduce an effective buffer training strategy (eBTS) that creates the optimized replay buffer on object detection. Our approach incorporates guarantee minimum and hierarchical sampling to establish the buffer customized to the trained model. %These methods can facilitate effective retrieval of prior knowledge. Furthermore, we use the circular experience replay training to optimally utilize the accumulated buffer data. Experiments on the MS COCO dataset demonstrate that our eBTS achieves state-of-the-art performance compared to the existing replay schemes.

Sungmook Woo, Hyungu Kang, Chanwoo Kim

Punctuation restoration improves ASR (Automatic Speech Recognition) readability. However streaming ASR requires online decisions with limited future context. In streaming ASR, the system predicts punctuation incrementally, which makes generation-based approaches prone to latency and alignment failures under boundary-wise evaluation. This paper proposes a non-autoregressive scoring method (no free-form generation) that preserves the input transcript and makes a decision at each word boundary. Our method compares punctuation insertion hypotheses against a no-insertion baseline under a bounded K-subword-token lookahead, and calibrates decisions using a weight α and a validation-calibrated threshold τ (no parameter updates during inference). On IWSLT 2017, our scoring method achieves a 4-class macro F1 of 0.893 in the no fine-tuning setting (validation-calibrated, K=2) and 0.937 after fine-tuning (K=2), outperforming the prompt-based baseline (0.566) and a fine-tuned ELECTRA baseline (0.913) under the same lookahead budget. We analyze the impact of the lookahead budget through ablation studies on K.

Chanwoo Park, Anna Seo Gyeong Choi, Sunghye Cho et al.

Societies worldwide are rapidly entering a super-aged era, making elderly health a pressing concern. The aging population is increasing the burden on national economies and households. Dementia cases are rising significantly with this demographic shift. Recent research using voice-based models and large language models (LLM) offers new possibilities for dementia diagnosis and treatment. Our Chain-of-Thought (CoT) reasoning method combines speech and language models. The process starts with automatic speech recognition to convert speech to text. We add a linear layer to an LLM for Alzheimer's disease (AD) and non-AD classification, using supervised fine-tuning (SFT) with CoT reasoning and cues. This approach showed an 16.7% relative performance improvement compared to methods without CoT prompt reasoning. To the best of our knowledge, our proposed method achieved state-of-the-art performance in CoT approaches.

Hadam Baek, Hannie Shin, Jiyoung Seo et al.

Accurately modeling sound propagation with complex real-world environments is essential for Novel View Acoustic Synthesis (NVAS). While previous studies have leveraged visual perception to estimate spatial acoustics, the combined use of surface normal and structural details from 3D representations in acoustic modeling has been underexplored. Given their direct impact on sound wave reflections and propagation, surface normals should be jointly modeled with structural details to achieve accurate spatial acoustics. In this paper, we propose a surface-enhanced geometry-aware approach for NVAS to improve spatial acoustic modeling. To achieve this, we exploit geometric priors, such as image, depth map, surface normals, and point clouds obtained using a 3D Gaussian Splatting (3DGS) based framework. We introduce a dual cross-attention-based transformer integrating geometrical constraints into frequency query to understand the surroundings of the emitter. Additionally, we design a ConvNeXt-based spectral features processing network called Spectral Refinement Network (SRN) to synthesize realistic binaural audio. Experimental results on the RWAVS and SoundSpace datasets highlight the necessity of our approach, as it surpasses existing methods in novel view acoustic synthesis.

Ireh Kim, Tesia Sker, Chanwoo Kim

In Machine Translation, Large Language Models (LLMs) have generally underperformed compared to conventional encoder-decoder systems and thus see limited adoption. However, LLMs excel at modeling contextual information, making them a natural fit for document-level translation tasks where coherence across sentences is crucial. Despite this potential, document-level MT with LLMs faces two key challenges: (1) the scarcity of large-scale, high-quality document-level parallel data; and (2) the propensity of LLMs to introduce hallucinations and omissions during generation. To address these challenges, we propose a two-stage fine-tuning strategy leveraging LLM-augmented document-level data. First, we augment data by converting summarization data into document-level parallel data using a LLM, and then filter it using multiple metrics, leveraging sacreBLEU, COMET, and LaBSE-based cosine similarity-to improve data quality. Finally, we employ a two-stage fine-tuning strategy: first fine-tuning on the abundant sentence-level MT resources, and then on the filtered document-level corpus.

Emily Alsentzer, Marie-Laure Charpignon, Bill Chen et al.

The 6th Annual Conference on Health, Inference, and Learning (CHIL 2025), hosted by the Association for Health Learning and Inference (AHLI), was held in person on June 25-27, 2025, at the University of California, Berkeley, in Berkeley, California, USA. As part of this year's program, we hosted Research Roundtables to catalyze collaborative, small-group dialogue around critical, timely topics at the intersection of machine learning and healthcare. Each roundtable was moderated by a team of senior and junior chairs who fostered open exchange, intellectual curiosity, and inclusive engagement. The sessions emphasized rigorous discussion of key challenges, exploration of emerging opportunities, and collective ideation toward actionable directions in the field. In total, eight roundtables were held by 19 roundtable chairs on topics of "Explainability, Interpretability, and Transparency," "Uncertainty, Bias, and Fairness," "Causality," "Domain Adaptation," "Foundation Models," "Learning from Small Medical Data," "Multimodal Methods," and "Scalable, Translational Healthcare Solutions."

Yerin Ryu, Inseop Shin, Chanwoo Kim

Controllable Singing Voice Synthesis (SVS) aims to generate expressive singing voices reflecting user intent. While recent SVS systems achieve high audio quality, most rely on probabilistic modeling, limiting precise control over attributes such as dynamics. We address this by focusing on dynamic control--temporal loudness variation essential for musical expressiveness--and explicitly condition the SVS model on energy sequences extracted from ground-truth spectrograms, reducing annotation costs and improving controllability. We also propose a phoneme-level energy sequence for user-friendly control. To the best of our knowledge, this is the first attempt enabling user-driven dynamics control in SVS. Experiments show our method achieves over 50% reduction in mean absolute error of energy sequences for phoneme-level inputs compared to baseline and energy-predictor models, without compromising synthesis quality.

Donggyun Kim, Chanwoo Kim, Seunghoon Hong

While test-time fine-tuning is beneficial in few-shot learning, the need for multiple backpropagation steps can be prohibitively expensive in real-time or low-resource scenarios. To address this limitation, we propose an approach that emulates gradient descent without computing gradients, enabling efficient test-time adaptation. Specifically, we formulate gradient descent as an Euler discretization of an ordinary differential equation (ODE) and train an auxiliary network to predict the task-conditional drift using only the few-shot support set. The adaptation then reduces to a simple numerical integration (e.g., via the Euler method), which requires only a few forward passes of the auxiliary network -- no gradients or forward passes of the target model are needed. In experiments on cross-domain few-shot classification using the Meta-Dataset and CDFSL benchmarks, our method significantly improves out-of-domain performance over the non-fine-tuned baseline while incurring only 6\% of the memory cost and 0.02\% of the computation time of standard fine-tuning, thus establishing a practical middle ground between direct transfer and fully fine-tuned approaches.

Hansoo Park, Chanwoo Kim, Jihyeon Kim et al.

RGB-based 3D pose estimation methods have been successful with the development of deep learning and the emergence of high-quality 3D pose datasets. However, most existing methods do not operate well for testing images whose distribution is far from that of training data. However, most existing methods do not operate well for testing images whose distribution is far from that of training data. This problem might be alleviated by involving diverse data during training, however it is non-trivial to collect such diverse data with corresponding labels (i.e. 3D pose). In this paper, we introduced an unsupervised domain adaptation framework for 3D pose estimation that utilizes the unlabeled data in addition to labeled data via masked image modeling (MIM) framework. Foreground-centric reconstruction and attention regularization are further proposed to increase the effectiveness of unlabeled data usage. Experiments are conducted on the various datasets in human and hand pose estimation tasks, especially using the cross-domain scenario. We demonstrated the effectiveness of ours by achieving the state-of-the-art accuracy on all datasets.

Chris Lin, Mingyu Lu, Chanwoo Kim et al.

As diffusion models are deployed in real-world settings, and their performance is driven by training data, appraising the contribution of data contributors is crucial to creating incentives for sharing quality data and to implementing policies for data compensation. Depending on the use case, model performance corresponds to various global properties of the distribution learned by a diffusion model (e.g., overall aesthetic quality). Hence, here we address the problem of attributing global properties of diffusion models to data contributors. The Shapley value provides a principled approach to valuation by uniquely satisfying game-theoretic axioms of fairness. However, estimating Shapley values for diffusion models is computationally impractical because it requires retraining on many training data subsets corresponding to different contributors and rerunning inference. We introduce a method to efficiently retrain and rerun inference for Shapley value estimation, by leveraging model pruning and fine-tuning. We evaluate the utility of our method with three use cases: (i) image quality for a DDPM trained on a CIFAR dataset, (ii) demographic diversity for an LDM trained on CelebA-HQ, and (iii) aesthetic quality for a Stable Diffusion model LoRA-finetuned on Post-Impressionist artworks. Our results empirically demonstrate that our framework can identify important data contributors across models' global properties, outperforming existing attribution methods for diffusion models.

Abhinav Garg, Jiyeon Kim, Sushil Khyalia et al.

Grapheme-to-Phoneme (G2P) is an essential first step in any modern, high-quality Text-to-Speech (TTS) system. Most of the current G2P systems rely on carefully hand-crafted lexicons developed by experts. This poses a two-fold problem. Firstly, the lexicons are generated using a fixed phoneme set, usually, ARPABET or IPA, which might not be the most optimal way to represent phonemes for all languages. Secondly, the man-hours required to produce such an expert lexicon are very high. In this paper, we eliminate both of these issues by using recent advances in self-supervised learning to obtain data-driven phoneme representations instead of fixed representations. We compare our lexicon-free approach against strong baselines that utilize a well-crafted lexicon. Furthermore, we show that our data-driven lexicon-free method performs as good or even marginally better than the conventional rule-based or lexicon-based neural G2Ps in terms of Mean Opinion Score (MOS) while using no prior language lexicon or phoneme set, i.e. no linguistic expertise.

Nauman Dawalatabad, Tushar Vatsal, Ashutosh Gupta et al.

Speech recognition on smart devices is challenging owing to the small memory footprint. Hence small size ASR models are desirable. With the use of popular transducer-based models, it has become possible to practically deploy streaming speech recognition models on small devices [1]. Recently, the two-pass model [2] combining RNN-T and LAS modules has shown exceptional performance for streaming on-device speech recognition. In this work, we propose a simple and effective approach to reduce the size of the two-pass model for memory-constrained devices. We employ a popular knowledge distillation approach in three stages using the Teacher-Student training technique. In the first stage, we use a trained RNN-T model as a teacher model and perform knowledge distillation to train the student RNN-T model. The second stage uses the shared encoder and trains a LAS rescorer for student model using the trained RNN-T+LAS teacher model. Finally, we perform deep-finetuning for the student model with a shared RNN-T encoder, RNN-T decoder, and LAS rescorer. Our experimental results on standard LibriSpeech dataset show that our system can achieve a high compression rate of 55% without significant degradation in the WER compared to the two-pass teacher model.

Jiyeon Kim, Mehul Kumar, Dhananjaya Gowda et al.

In this paper, we propose a three-stage training methodology to improve the speech recognition accuracy of low-resource languages. We explore and propose an effective combination of techniques such as transfer learning, encoder freezing, data augmentation using Text-To-Speech (TTS), and Semi-Supervised Learning (SSL). To improve the accuracy of a low-resource Italian ASR, we leverage a well-trained English model, unlabeled text corpus, and unlabeled audio corpus using transfer learning, TTS augmentation, and SSL respectively. In the first stage, we use transfer learning from a well-trained English model. This primarily helps in learning the acoustic information from a resource-rich language. This stage achieves around 24% relative Word Error Rate (WER) reduction over the baseline. In stage two, We utilize unlabeled text data via TTS data-augmentation to incorporate language information into the model. We also explore freezing the acoustic encoder at this stage. TTS data augmentation helps us further reduce the WER by ~ 21% relatively. Finally, In stage three we reduce the WER by another 4% relative by using SSL from unlabeled audio data. Overall, our two-pass speech recognition system with a Monotonic Chunkwise Attention (MoChA) in the first pass and a full-attention in the second pass achieves a WER reduction of ~ 42% relative to the baseline.

Jiyeon Kim, Mehul Kumar, Dhananjaya Gowda et al.

In this paper, we present a comparative study on the robustness of two different online streaming speech recognition models: Monotonic Chunkwise Attention (MoChA) and Recurrent Neural Network-Transducer (RNN-T). We explore three recently proposed data augmentation techniques, namely, multi-conditioned training using an acoustic simulator, Vocal Tract Length Perturbation (VTLP) for speaker variability, and SpecAugment. Experimental results show that unidirectional models are in general more sensitive to noisy examples in the training set. It is observed that the final performance of the model depends on the proportion of training examples processed by data augmentation techniques. MoChA models generally perform better than RNN-T models. However, we observe that training of MoChA models seems to be more sensitive to various factors such as the characteristics of training sets and the incorporation of additional augmentations techniques. On the other hand, RNN-T models perform better than MoChA models in terms of latency, inference time, and the stability of training. Additionally, RNN-T models are generally more robust against noise and reverberation. All these advantages make RNN-T models a better choice for streaming on-device speech recognition compared to MoChA models.

Mohd Abbas Zaidi, Beomseok Lee, Sangha Kim et al.

Simultaneous translation systems start producing the output while processing the partial source sentence in the incoming input stream. These systems need to decide when to read more input and when to write the output. These decisions depend on the structure of source/target language and the information contained in the partial input sequence. Hence, read/write decision policy remains the same across different input modalities, i.e., speech and text. This motivates us to leverage the text transcripts corresponding to the speech input for improving simultaneous speech-to-text translation (SimulST). We propose Decision Attentive Regularization (DAR) to improve the decision policy of SimulST systems by using the simultaneous text-to-text translation (SimulMT) task. We also extend several techniques from the offline speech translation domain to explore the role of SimulMT task in improving SimulST performance. Overall, we achieve 34.66% / 4.5 BLEU improvement over the baseline model across different latency regimes for the MuST-C English-German (EnDe) SimulST task.

Chanwoo Kim, Abhinav Garg, Dhananjaya Gowda et al.

In this paper, we present a streaming end-to-end speech recognition model based on Monotonic Chunkwise Attention (MoCha) jointly trained with enhancement layers. Even though the MoCha attention enables streaming speech recognition with recognition accuracy comparable to a full attention-based approach, training this model is sensitive to various factors such as the difficulty of training examples, hyper-parameters, and so on. Because of these issues, speech recognition accuracy of a MoCha-based model for clean speech drops significantly when a multi-style training approach is applied. Inspired by Curriculum Learning [1], we introduce two training strategies: Gradual Application of Enhanced Features (GAEF) and Gradual Reduction of Enhanced Loss (GREL). With GAEF, the model is initially trained using clean features. Subsequently, the portion of outputs from the enhancement layers gradually increases. With GREL, the portion of the Mean Squared Error (MSE) loss for the enhanced output gradually reduces as training proceeds. In experimental results on the LibriSpeech corpus and noisy far-field test sets, the proposed model with GAEF-GREL training strategies shows significantly better results than the conventional multi-style training approach.

Hyojung Han, Sathish Indurthi, Mohd Abbas Zaidi et al.

Recently, simultaneous translation has gathered a lot of attention since it enables compelling applications such as subtitle translation for a live event or real-time video-call translation. Some of these translation applications allow editing of partial translation giving rise to re-translation approaches. The current re-translation approaches are based on autoregressive sequence generation models (ReTA), which generate tar-get tokens in the (partial) translation sequentially. The multiple re-translations with sequential generation inReTAmodelslead to an increased inference time gap between the incoming source input and the corresponding target output as the source input grows. Besides, due to the large number of inference operations involved, the ReTA models are not favorable for resource-constrained devices. In this work, we propose a faster re-translation system based on a non-autoregressive sequence generation model (FReTNA) to overcome the aforementioned limitations. We evaluate the proposed model on multiple translation tasks and our model reduces the inference times by several orders and achieves a competitive BLEUscore compared to the ReTA and streaming (Wait-k) models.The proposed model reduces the average computation time by a factor of 20 when compared to the ReTA model by incurring a small drop in the translation quality. It also outperforms the streaming-based Wait-k model both in terms of computation time (1.5 times lower) and translation quality.

Chanwoo Kim, Dhananjaya Gowda, Dongsoo Lee et al.

In this paper, we review various end-to-end automatic speech recognition algorithms and their optimization techniques for on-device applications. Conventional speech recognition systems comprise a large number of discrete components such as an acoustic model, a language model, a pronunciation model, a text-normalizer, an inverse-text normalizer, a decoder based on a Weighted Finite State Transducer (WFST), and so on. To obtain sufficiently high speech recognition accuracy with such conventional speech recognition systems, a very large language model (up to 100 GB) is usually needed. Hence, the corresponding WFST size becomes enormous, which prohibits their on-device implementation. Recently, fully neural network end-to-end speech recognition algorithms have been proposed. Examples include speech recognition systems based on Connectionist Temporal Classification (CTC), Recurrent Neural Network Transducer (RNN-T), Attention-based Encoder-Decoder models (AED), Monotonic Chunk-wise Attention (MoChA), transformer-based speech recognition systems, and so on. These fully neural network-based systems require much smaller memory footprints compared to conventional algorithms, therefore their on-device implementation has become feasible. In this paper, we review such end-to-end speech recognition models. We extensively discuss their structures, performance, and advantages compared to conventional algorithms.

Chanwoo Kim, Kwangyoun Kim, Sathish Reddy Indurthi

In this paper, we present a Small Energy Masking (SEM) algorithm, which masks inputs having values below a certain threshold. More specifically, a time-frequency bin is masked if the filterbank energy in this bin is less than a certain energy threshold. A uniform distribution is employed to randomly generate the ratio of this energy threshold to the peak filterbank energy of each utterance in decibels. The unmasked feature elements are scaled so that the total sum of the feature values remain the same through this masking procedure. This very simple algorithm shows relatively 11.2 % and 13.5 % Word Error Rate (WER) improvements on the standard LibriSpeech test-clean and test-other sets over the baseline end-to-end speech recognition system. Additionally, compared to the input dropout algorithm, SEM algorithm shows relatively 7.7 % and 11.6 % improvements on the same LibriSpeech test-clean and test-other sets. With a modified shallow-fusion technique with a Transformer LM, we obtained a 2.62 % WER on the LibriSpeech test-clean set and a 7.87 % WER on the LibriSpeech test-other set.

Kwangyoun Kim, Kyungmin Lee, Dhananjaya Gowda et al.

In this paper, we present a new on-device automatic speech recognition (ASR) system based on monotonic chunk-wise attention (MoChA) models trained with large (> 10K hours) corpus. We attained around 90% of a word recognition rate for general domain mainly by using joint training of connectionist temporal classifier (CTC) and cross entropy (CE) losses, minimum word error rate (MWER) training, layer-wise pre-training and data augmentation methods. In addition, we compressed our models by more than 3.4 times smaller using an iterative hyper low-rank approximation (LRA) method while minimizing the degradation in recognition accuracy. The memory footprint was further reduced with 8-bit quantization to bring down the final model size to lower than 39 MB. For on-demand adaptation, we fused the MoChA models with statistical n-gram models, and we could achieve a relatively 36% improvement on average in word error rate (WER) for target domains including the general domain.

Abhinav Garg, Dhananjaya Gowda, Ankur Kumar et al.

In this paper, we propose a refined multi-stage multi-task training strategy to improve the performance of online attention-based encoder-decoder (AED) models. A three-stage training based on three levels of architectural granularity namely, character encoder, byte pair encoding (BPE) based encoder, and attention decoder, is proposed. Also, multi-task learning based on two-levels of linguistic granularity namely, character and BPE, is used. We explore different pre-training strategies for the encoders including transfer learning from a bidirectional encoder. Our encoder-decoder models with online attention show 35% and 10% relative improvement over their baselines for smaller and bigger models, respectively. Our models achieve a word error rate (WER) of 5.04% and 4.48% on the Librispeech test-clean data for the smaller and bigger models respectively after fusion with long short-term memory (LSTM) based external language model (LM).

Chanwoo Kim, Mehul Kumar, Kwangyoun Kim et al.

In this paper, we describe the Maximum Uniformity of Distribution (MUD) algorithm with the power-law nonlinearity. In this approach, we hypothesize that neural network training will become more stable if feature distribution is not too much skewed. We propose two different types of MUD approaches: power function-based MUD and histogram-based MUD. In these approaches, we first obtain the mel filterbank coefficients and apply nonlinearity functions for each filterbank channel. With the power function-based MUD, we apply a power-function based nonlinearity where power function coefficients are chosen to maximize the likelihood assuming that nonlinearity outputs follow the uniform distribution. With the histogram-based MUD, the empirical Cumulative Density Function (CDF) from the training database is employed to transform the original distribution into a uniform distribution. In MUD processing, we do not use any prior knowledge (e.g. logarithmic relation) about the energy of the incoming signal and the perceived intensity by a human. Experimental results using an end-to-end speech recognition system demonstrate that power-function based MUD shows better result than the conventional Mel Filterbank Cepstral Coefficients (MFCCs). On the LibriSpeech database, we could achieve 4.02 % WER on test-clean and 13.34 % WER on test-other without using any Language Models (LMs). The major contribution of this work is that we developed a new algorithm for designing the compressive nonlinearity in a data-driven way, which is much more flexible than the previous approaches and may be extended to other domains as well.

Chanwoo Kim, Sungsoo Kim, Kwangyoun Kim et al.

In this paper, we present an end-to-end training framework for building state-of-the-art end-to-end speech recognition systems. Our training system utilizes a cluster of Central Processing Units(CPUs) and Graphics Processing Units (GPUs). The entire data reading, large scale data augmentation, neural network parameter updates are all performed "on-the-fly". We use vocal tract length perturbation [1] and an acoustic simulator [2] for data augmentation. The processed features and labels are sent to the GPU cluster. The Horovod allreduce approach is employed to train neural network parameters. We evaluated the effectiveness of our system on the standard Librispeech corpus [3] and the 10,000-hr anonymized Bixby English dataset. Our end-to-end speech recognition system built using this training infrastructure showed a 2.44 % WER on test-clean of the LibriSpeech test set after applying shallow fusion with a Transformer language model (LM). For the proprietary English Bixby open domain test set, we obtained a WER of 7.92 % using a Bidirectional Full Attention (BFA) end-to-end model after applying shallow fusion with an RNN-LM. When the monotonic chunckwise attention (MoCha) based approach is employed for streaming speech recognition, we obtained a WER of 9.95 % on the same Bixby open domain test set.

Sathish Indurthi, Houjeung Han, Nikhil Kumar Lakumarapu et al.

End-to-end Speech Translation (ST) models have several advantages such as lower latency, smaller model size, and less error compounding over conventional pipelines that combine Automatic Speech Recognition (ASR) and text Machine Translation (MT) models. However, collecting large amounts of parallel data for ST task is more difficult compared to the ASR and MT tasks. Previous studies have proposed the use of transfer learning approaches to overcome the above difficulty. These approaches benefit from weakly supervised training data, such as ASR speech-to-transcript or MT text-to-text translation pairs. However, the parameters in these models are updated independently of each task, which may lead to sub-optimal solutions. In this work, we adopt a meta-learning algorithm to train a modality agnostic multi-task model that transfers knowledge from source tasks=ASR+MT to target task=ST where ST task severely lacks data. In the meta-learning phase, the parameters of the model are exposed to vast amounts of speech transcripts (e.g., English ASR) and text translations (e.g., English-German MT). During this phase, parameters are updated in such a way to understand speech, text representations, the relation between them, as well as act as a good initialization point for the target ST task. We evaluate the proposed meta-learning approach for ST tasks on English-German (En-De) and English-French (En-Fr) language pairs from the Multilingual Speech Translation Corpus (MuST-C). Our method outperforms the previous transfer learning approaches and sets new state-of-the-art results for En-De and En-Fr ST tasks by obtaining 9.18, and 11.76 BLEU point improvements, respectively.