Jian Xue

Jian Xue, Peidong Wang, Jinyu Li et al. · microsoft-research

Neural transducers have been widely used in automatic speech recognition (ASR). In this paper, we introduce it to streaming end-to-end speech translation (ST), which aims to convert audio signals to texts in other languages directly. Compared with cascaded ST that performs ASR followed by text-based machine translation (MT), the proposed Transformer transducer (TT)-based ST model drastically reduces inference latency, exploits speech information, and avoids error propagation from ASR to MT. To improve the modeling capacity, we propose attention pooling for the joint network in TT. In addition, we extend TT-based ST to multilingual ST, which generates texts of multiple languages at the same time. Experimental results on a large-scale 50 thousand (K) hours pseudo-labeled training set show that TT-based ST not only significantly reduces inference time but also outperforms non-streaming cascaded ST for English-German translation.

Mu Yang, Naoyuki Kanda, Xiaofei Wang et al. · cmu

End-to-end speech translation (ST) for conversation recordings involves several under-explored challenges such as speaker diarization (SD) without accurate word time stamps and handling of overlapping speech in a streaming fashion. In this work, we propose DiariST, the first streaming ST and SD solution. It is built upon a neural transducer-based streaming ST system and integrates token-level serialized output training and t-vector, which were originally developed for multi-talker speech recognition. Due to the absence of evaluation benchmarks in this area, we develop a new evaluation dataset, DiariST-AliMeeting, by translating the reference Chinese transcriptions of the AliMeeting corpus into English. We also propose new metrics, called speaker-agnostic BLEU and speaker-attributed BLEU, to measure the ST quality while taking SD accuracy into account. Our system achieves a strong ST and SD capability compared to offline systems based on Whisper, while performing streaming inference for overlapping speech. To facilitate the research in this new direction, we release the evaluation data, the offline baseline systems, and the evaluation code.

Xing Lan, Jiayi Lyu, Hanyu Jiang et al.

In this paper, we explore the zero-shot capability of the Segment Anything Model (SAM) for food image segmentation. To address the lack of class-specific information in SAM-generated masks, we propose a novel framework, called FoodSAM. This innovative approach integrates the coarse semantic mask with SAM-generated masks to enhance semantic segmentation quality. Besides, we recognize that the ingredients in food can be supposed as independent individuals, which motivated us to perform instance segmentation on food images. Furthermore, FoodSAM extends its zero-shot capability to encompass panoptic segmentation by incorporating an object detector, which renders FoodSAM to effectively capture non-food object information. Drawing inspiration from the recent success of promptable segmentation, we also extend FoodSAM to promptable segmentation, supporting various prompt variants. Consequently, FoodSAM emerges as an all-encompassing solution capable of segmenting food items at multiple levels of granularity. Remarkably, this pioneering framework stands as the first-ever work to achieve instance, panoptic, and promptable segmentation on food images. Extensive experiments demonstrate the feasibility and impressing performance of FoodSAM, validating SAM's potential as a prominent and influential tool within the domain of food image segmentation. We release our code at https://github.com/jamesjg/FoodSAM.

Liping Hou, Ke Lu, Xue Yang et al.

Typical representations for arbitrary-oriented object detection tasks include oriented bounding box (OBB), quadrilateral bounding box (QBB), and point set (PointSet). Each representation encounters problems that correspond to its characteristics, such as the boundary discontinuity, square-like problem, representation ambiguity, and isolated points, which lead to inaccurate detection. Although many effective strategies have been proposed for various representations, there is still no unified solution. Current detection methods based on Gaussian modeling have demonstrated the possibility of breaking this dilemma; however, they remain limited to OBB. To go further, in this paper, we propose a unified Gaussian representation called G-Rep to construct Gaussian distributions for OBB, QBB, and PointSet, which achieves a unified solution to various representations and problems. Specifically, PointSet or QBB-based object representations are converted into Gaussian distributions, and their parameters are optimized using the maximum likelihood estimation algorithm. Then, three optional Gaussian metrics are explored to optimize the regression loss of the detector because of their excellent parameter optimization mechanisms. Furthermore, we also use Gaussian metrics for sampling to align label assignment and regression loss. Experimental results on several public available datasets, such as DOTA, HRSC2016, UCAS-AOD, and ICDAR2015, show the excellent performance of the proposed method for arbitrary-oriented object detection.

Rui Zhao, Jian Xue, Partha Parthasarathy et al.

Neural transducer is now the most popular end-to-end model for speech recognition, due to its naturally streaming ability. However, it is challenging to adapt it with text-only data. Factorized neural transducer (FNT) model was proposed to mitigate this problem. The improved adaptation ability of FNT on text-only adaptation data came at the cost of lowered accuracy compared to the standard neural transducer model. We propose several methods to improve the performance of the FNT model. They are: adding CTC criterion during training, adding KL divergence loss during adaptation, using a pre-trained language model to seed the vocabulary predictor, and an efficient adaptation approach by interpolating the vocabulary predictor with the n-gram language model. A combination of these approaches results in a relative word-error-rate reduction of 9.48\% from the standard FNT model. Furthermore, n-gram interpolation with the vocabulary predictor improves the adaptation speed hugely with satisfactory adaptation performance.

Jian Xue, Peidong Wang, Jinyu Li et al.

In this paper, we introduce our work of building a Streaming Multilingual Speech Model (SM2), which can transcribe or translate multiple spoken languages into texts of the target language. The backbone of SM2 is Transformer Transducer, which has high streaming capability. Instead of human labeled speech translation (ST) data, SM2 models are trained using weakly supervised data generated by converting the transcriptions in speech recognition corpora with a machine translation service. With 351 thousand hours of anonymized speech training data from 25 languages, SM2 models achieve comparable or even better ST quality than some recent popular large-scale non-streaming speech models. More importantly, we show that SM2 has the truly zero-shot capability when expanding to new target languages, yielding high quality ST results for {source-speech, target-text} pairs that are not seen during training.

Peidong Wang, Eric Sun, Jian Xue et al.

Automatic speech recognition (ASR) and speech translation (ST) can both use neural transducers as the model structure. It is thus possible to use a single transducer model to perform both tasks. In real-world applications, such joint ASR and ST models may need to be streaming and do not require source language identification (i.e. language-agnostic). In this paper, we propose LAMASSU, a streaming language-agnostic multilingual speech recognition and translation model using neural transducers. Based on the transducer model structure, we propose four methods, a unified joint and prediction network for multilingual output, a clustered multilingual encoder, target language identification for encoder, and connectionist temporal classification regularization. Experimental results show that LAMASSU not only drastically reduces the model size but also reaches the performances of monolingual ASR and bilingual ST models.

Sara Papi, Peidong Wang, Junkun Chen et al.

In real-world applications, users often require both translations and transcriptions of speech to enhance their comprehension, particularly in streaming scenarios where incremental generation is necessary. This paper introduces a streaming Transformer-Transducer that jointly generates automatic speech recognition (ASR) and speech translation (ST) outputs using a single decoder. To produce ASR and ST content effectively with minimal latency, we propose a joint token-level serialized output training method that interleaves source and target words by leveraging an off-the-shelf textual aligner. Experiments in monolingual (it-en) and multilingual (\{de,es,it\}-en) settings demonstrate that our approach achieves the best quality-latency balance. With an average ASR latency of 1s and ST latency of 1.3s, our model shows no degradation or even improves output quality compared to separate ASR and ST models, yielding an average improvement of 1.1 WER and 0.4 BLEU in the multilingual case.

Eric Sun, Jinyu Li, Yuxuan Hu et al.

We propose gated language experts and curriculum training to enhance multilingual transformer transducer models without requiring language identification (LID) input from users during inference. Our method incorporates a gating mechanism and LID loss, enabling transformer experts to learn language-specific information. By combining gated transformer experts with shared transformer layers, we construct multilingual transformer blocks and utilize linear experts to effectively regularize the joint network. The curriculum training scheme leverages LID to guide the gated experts in improving their respective language performance. Experimental results on a bilingual task involving English and Spanish demonstrate significant improvements, with average relative word error reductions of 12.5% and 7.3% compared to the baseline bilingual and monolingual models, respectively. Notably, our method achieves performance comparable to the upper-bound model trained and inferred with oracle LID. Extending our approach to trilingual, quadrilingual, and pentalingual models reveals similar advantages to those observed in the bilingual models, highlighting its ease of extension to multiple languages.

Yashesh Gaur, Nick Kibre, Jian Xue et al.

Automatic Speech Recognition (ASR) systems typically yield output in lexical form. However, humans prefer a written form output. To bridge this gap, ASR systems usually employ Inverse Text Normalization (ITN). In previous works, Weighted Finite State Transducers (WFST) have been employed to do ITN. WFSTs are nicely suited to this task but their size and run-time costs can make deployment on embedded applications challenging. In this paper, we describe the development of an on-device ITN system that is streaming, lightweight & accurate. At the core of our system is a streaming transformer tagger, that tags lexical tokens from ASR. The tag informs which ITN category might be applied, if at all. Following that, we apply an ITN-category-specific WFST, only on the tagged text, to reliably perform the ITN conversion. We show that the proposed ITN solution performs equivalent to strong baselines, while being significantly smaller in size and retaining customization capabilities.

Sara Papi, Peidong Wang, Junkun Chen et al.

The growing need for instant spoken language transcription and translation is driven by increased global communication and cross-lingual interactions. This has made offering translations in multiple languages essential for user applications. Traditional approaches to automatic speech recognition (ASR) and speech translation (ST) have often relied on separate systems, leading to inefficiencies in computational resources, and increased synchronization complexity in real time. In this paper, we propose a streaming Transformer-Transducer (T-T) model able to jointly produce many-to-one and one-to-many transcription and translation using a single decoder. We introduce a novel method for joint token-level serialized output training based on timestamp information to effectively produce ASR and ST outputs in the streaming setting. Experiments on {it,es,de}->en prove the effectiveness of our approach, enabling the generation of one-to-many joint outputs with a single decoder for the first time.

Jinbao Wang, Ke Lu, Jian Xue

This paper proposes a novel application system for the generation of three-dimensional (3D) character animation driven by markerless human body motion capturing. The entire pipeline of the system consists of five stages: 1) the capturing of motion data using multiple cameras, 2) detection of the two-dimensional (2D) human body joints, 3) estimation of the 3D joints, 4) calculation of bone transformation matrices, and 5) generation of character animation. The main objective of this study is to generate a 3D skeleton and animation for 3D characters using multi-view images captured by ordinary cameras. The computational complexity of the 3D skeleton reconstruction based on 3D vision has been reduced as needed to achieve frame-by-frame motion capturing. The experimental results reveal that our system can effectively and efficiently capture human actions and use them to animate 3D cartoon characters in real-time.

Xing Lan, Jian Xue, Ji Qi et al.

Facial expression recognition (FER) is a critical task in multimedia with significant implications across various domains. However, analyzing the causes of facial expressions is essential for accurately recognizing them. Current approaches, such as those based on facial action units (AUs), typically provide AU names and intensities but lack insight into the interactions and relationships between AUs and the overall expression. In this paper, we propose a novel method called ExpLLM, which leverages large language models to generate an accurate chain of thought (CoT) for facial expression recognition. Specifically, we have designed the CoT mechanism from three key perspectives: key observations, overall emotional interpretation, and conclusion. The key observations describe the AU's name, intensity, and associated emotions. The overall emotional interpretation provides an analysis based on multiple AUs and their interactions, identifying the dominant emotions and their relationships. Finally, the conclusion presents the final expression label derived from the preceding analysis. Furthermore, we also introduce the Exp-CoT Engine, designed to construct this expression CoT and generate instruction-description data for training our ExpLLM. Extensive experiments on the RAF-DB and AffectNet datasets demonstrate that ExpLLM outperforms current state-of-the-art FER methods. ExpLLM also surpasses the latest GPT-4o in expression CoT generation, particularly in recognizing micro-expressions where GPT-4o frequently fails.

Guohong Hu, Xing Lan, Hanyu Jiang et al.

Facial Action Units (AUs) are of great significance in the realm of affective computing. In this paper, we propose AU-LLaVA, the first unified AU recognition framework based on the Large Language Model (LLM). AU-LLaVA consists of a visual encoder, a linear projector layer, and a pre-trained LLM. We meticulously craft the text descriptions and fine-tune the model on various AU datasets, allowing it to generate different formats of AU recognition results for the same input image. On the BP4D and DISFA datasets, AU-LLaVA delivers the most accurate recognition results for nearly half of the AUs. Our model achieves improvements of F1-score up to 11.4% in specific AU recognition compared to previous benchmark results. On the FEAFA dataset, our method achieves significant improvements over all 24 AUs compared to previous benchmark results. AU-LLaVA demonstrates exceptional performance and versatility in AU recognition.

Tong Ning, Ke Lu, Xirui Jiang et al.

Utilizing temporal information to improve the performance of 3D detection has made great progress recently in the field of autonomous driving. Traditional transformer-based temporal fusion methods suffer from quadratic computational cost and information decay as the length of the frame sequence increases. In this paper, we propose a novel method called MambaDETR, whose main idea is to implement temporal fusion in the efficient state space. Moreover, we design a Motion Elimination module to remove the relatively static objects for temporal fusion. On the standard nuScenes benchmark, our proposed MambaDETR achieves remarkable result in the 3D object detection task, exhibiting state-of-the-art performance among existing temporal fusion methods.

Harveen Singh Chadha, Aswin Shanmugam Subramanian, Vikas Joshi et al.

In video dubbing, aligning translated audio with the source audio is a significant challenge. Our focus is on achieving this efficiently, tailored for real-time, on-device video dubbing scenarios. We developed a phoneme-based end-to-end length-sensitive speech translation (LSST) model, which generates translations of varying lengths short, normal, and long using predefined tags. Additionally, we introduced length-aware beam search (LABS), an efficient approach to generate translations of different lengths in a single decoding pass. This approach maintained comparable BLEU scores compared to a baseline without length awareness while significantly enhancing synchronization quality between source and target audio, achieving a mean opinion score (MOS) gain of 0.34 for Spanish and 0.65 for Korean, respectively.

Peidong Wang, Jian Xue, Rui Zhao et al.

Phrases are essential to understand the core concepts in conversations. However, due to their rare occurrence in training data, correct translation of phrases is challenging in speech translation tasks. In this paper, we propose a phrase dictionary biasing method to leverage pairs of phrases mapping from the source language to the target language. We apply the phrase dictionary biasing method to two types of widely adopted models, a transducer-based streaming speech translation model and a multimodal large language model. Experimental results show that the phrase dictionary biasing method outperforms phrase list biasing by 21% relatively for the streaming speech translation model. In addition, phrase dictionary biasing enables multimodal large language models to use external phrase information, achieving 85% relative improvement in phrase recall.

Peidong Wang, Naoyuki Kanda, Jian Xue et al.

Streaming multi-talker speech translation is a task that involves not only generating accurate and fluent translations with low latency but also recognizing when a speaker change occurs and what the speaker's gender is. Speaker change information can be used to create audio prompts for a zero-shot text-to-speech system, and gender can help to select speaker profiles in a conventional text-to-speech model. We propose to tackle streaming speaker change detection and gender classification by incorporating speaker embeddings into a transducer-based streaming end-to-end speech translation model. Our experiments demonstrate that the proposed methods can achieve high accuracy for both speaker change detection and gender classification.

Peidong Wang, Jian Xue, Jinyu Li et al.

Language-agnostic many-to-one end-to-end speech translation models can convert audio signals from different source languages into text in a target language. These models do not need source language identification, which improves user experience. In some cases, the input language can be given or estimated. Our goal is to use this additional language information while preserving the quality of the other languages. We accomplish this by introducing a simple and effective linear input network. The linear input network is initialized as an identity matrix, which ensures that the model can perform as well as, or better than, the original model. Experimental results show that the proposed method can successfully enhance the specified language, while keeping the language-agnostic ability of the many-to-one ST models.

Wei Gan, Jian Xue, Ke Lu et al.

Nearly all existing Facial Action Coding System-based datasets that include facial action unit (AU) intensity information annotate the intensity values hierarchically using A--E levels. However, facial expressions change continuously and shift smoothly from one state to another. Therefore, it is more effective to regress the intensity value of local facial AUs to represent whole facial expression changes, particularly in the fields of expression transfer and facial animation. We introduce an extension of FEAFA in combination with the relabeled DISFA database, which is available at https://www.iiplab.net/feafa+/ now. Extended FEAFA (FEAFA+) includes 150 video sequences from FEAFA and DISFA, with a total of 230,184 frames being manually annotated on floating-point intensity value of 24 redefined AUs using the Expression Quantitative Tool. We also list crude numerical results for posed and spontaneous subsets and provide a baseline comparison for the AU intensity regression task.

Rui Zhao, Jian Xue, Jinyu Li et al.

In this paper, several works are proposed to address practical challenges for deploying RNN Transducer (RNN-T) based speech recognition system. These challenges are adapting a well-trained RNN-T model to a new domain without collecting the audio data, obtaining time stamps and confidence scores at word level. The first challenge is solved with a splicing data method which concatenates the speech segments extracted from the source domain data. To get the time stamp, a phone prediction branch is added to the RNN-T model by sharing the encoder for the purpose of force alignment. Finally, we obtain word-level confidence scores by utilizing several types of features calculated during decoding and from confusion network. Evaluated with Microsoft production data, the splicing data adaptation method improves the baseline and adaptation with the text to speech method by 58.03% and 15.25% relative word error rate reduction, respectively. The proposed time stamping method can get less than 50ms word timing difference from the ground truth alignment on average while maintaining the recognition accuracy of the RNN-T model. We also obtain high confidence annotation performance with limited computation cost.

Yanfu Yan, Ke Lu, Jian Xue et al.

Facial expression analysis based on machine learning requires large number of well-annotated data to reflect different changes in facial motion. Publicly available datasets truly help to accelerate research in this area by providing a benchmark resource, but all of these datasets, to the best of our knowledge, are limited to rough annotations for action units, including only their absence, presence, or a five-level intensity according to the Facial Action Coding System. To meet the need for videos labeled in great detail, we present a well-annotated dataset named FEAFA for Facial Expression Analysis and 3D Facial Animation. One hundred and twenty-two participants, including children, young adults and elderly people, were recorded in real-world conditions. In addition, 99,356 frames were manually labeled using Expression Quantitative Tool developed by us to quantify 9 symmetrical FACS action units, 10 asymmetrical (unilateral) FACS action units, 2 symmetrical FACS action descriptors and 2 asymmetrical FACS action descriptors, and each action unit or action descriptor is well-annotated with a floating point number between 0 and 1. To provide a baseline for use in future research, a benchmark for the regression of action unit values based on Convolutional Neural Networks are presented. We also demonstrate the potential of our FEAFA dataset for 3D facial animation. Almost all state-of-the-art algorithms for facial animation are achieved based on 3D face reconstruction. We hence propose a novel method that drives virtual characters only based on action unit value regression of the 2D video frames of source actors.