分享

Modern automatic speech recognition (ASR) model is required to accurately transcribe diverse speech signals (from different domains, languages, accents, etc) given the specific contextual information in various application scenarios. Classic end-to-end models fused with extra language models perform well, but mainly in data matching scenarios and are gradually approaching a bottleneck. In this work, we introduce Seed-ASR, a large language model (LLM) based speech recognition model. Seed-ASR is developed based on the framework of audio conditioned LLM (AcLLM), leveraging the capabilities of LLMs by inputting continuous speech representations together with contextual information into the LLM. Through stage-wise large-scale training and the elicitation of context-aware capabilities in LLM, Seed-ASR demonstrates significant improvement over end-to-end models on comprehensive evaluation sets, including multiple domains, accents/dialects and languages. Additionally, Seed-ASR can be further deployed to support specific needs in various scenarios without requiring extra language models. Compared to recently released large ASR models, Seed-ASR achieves 10%-40% reduction in word (or character, for Chinese) error rates on Chinese and English public test sets, further demonstrating its powerful performance.

Generative adversarial network (GAN) based vocoders have achieved significant attention in speech synthesis with high quality and fast inference speed. However, there still exist many noticeable spectral artifacts, resulting in the quality decline of synthesized speech. In this work, we adopt a novel GAN-based vocoder designed for few artifacts and high fidelity, called FA-GAN. To suppress the aliasing artifacts caused by non-ideal upsampling layers in high-frequency components, we introduce the anti-aliased twin deconvolution module in the generator. To alleviate blurring artifacts and enrich the reconstruction of spectral details, we propose a novel fine-grained multi-resolution real and imaginary loss to assist in the modeling of phase information. Experimental results reveal that FA-GAN outperforms the compared approaches in promoting audio quality and alleviating spectral artifacts, and exhibits superior performance when applied to unseen speaker scenarios.

Code-switching (CS) phenomenon occurs when words or phrases from different languages are alternated in a single sentence. Due to data scarcity, building an effective CS Automatic Speech Recognition (ASR) system remains challenging. In this paper, we propose to enhance CS-ASR systems by utilizing rich unsupervised monolingual speech data within a semi-supervised learning framework, particularly when access to CS data is limited. To achieve this, we establish a general paradigm for applying noisy student training (NST) to the CS-ASR task. Specifically, we introduce the LLM-Filter, which leverages well-designed prompt templates to activate the correction capability of large language models (LLMs) for monolingual data selection and pseudo-labels refinement during NST. Our experiments on the supervised ASRU-CS and unsupervised AISHELL-2 and LibriSpeech datasets show that our method not only achieves significant improvements over supervised and semi-supervised learning baselines for the CS task, but also attains better performance compared with the fully-supervised oracle upper-bound on the CS English part. Additionally, we further investigate the influence of accent on AESRC dataset and demonstrate that our method can get achieve additional benefits when the monolingual data contains relevant linguistic characteristic.

Automatic speaker verification (ASV) systems are vulnerable to spoofing attacks such as text-to-speech. In this study, we propose a novel spoofing-robust ASV back-end classifier, optimized directly for the recently introduced, architecture-agnostic detection cost function (a-DCF). We combine a-DCF and binary cross-entropy (BCE) losses to optimize the network weights, combined by a novel, straightforward detection threshold optimization technique. Experiments on the ASVspoof2019 database demonstrate considerable improvement over the baseline optimized using BCE only (from minimum a-DCF of 0.1445 to 0.1254), representing 13% relative improvement. These initial promising results demonstrate that it is possible to adjust an ASV system to find appropriate balance across the contradicting aims of user convenience and security against adversaries.

Training modern LLMs is extremely resource intensive, and customizing them for various deployment scenarios characterized by limited compute and memory resources through repeated training is impractical. In this paper, we introduce Flextron, a network architecture and post-training model optimization framework supporting flexible model deployment. The Flextron architecture utilizes a nested elastic structure to rapidly adapt to specific user-defined latency and accuracy targets during inference with no additional fine-tuning required. It is also input-adaptive, and can automatically route tokens through its sub-networks for improved performance and efficiency. We present a sample-efficient training method and associated routing algorithms for systematically transforming an existing trained LLM into a Flextron model. We evaluate Flextron on the GPT-3 and LLama-2 family of LLMs, and demonstrate superior performance over multiple end-to-end trained variants and other state-of-the-art elastic networks, all with a single pretraining run that consumes a mere 7.63% tokens compared to original pretraining.

We surely enjoy the larger the better models for their superior performance in the last couple of years when both the hardware and software support the birth of such extremely huge models. The applied fields include text mining and others. In particular, the success of LLMs on text understanding and text generation draws attention from researchers who have worked on NLP and related areas for years or even decades. On the side, LLMs may suffer from problems like model overfitting, hallucination, and device limitation to name a few. In this work, we suggest a model pruning technique specifically focused on LLMs. The proposed methodology emphasizes the explainability of deep learning models. By having the theoretical foundation, we obtain a trustworthy deep model so that huge models with a massive number of model parameters become not quite necessary. A mutual information-based estimation is adopted to find neurons with redundancy to eliminate. Moreover, an estimator with well-tuned parameters helps to find precise estimation to guide the pruning procedure. At the same time, we also explore the difference between pruning on large-scale models vs. pruning on small-scale models. The choice of pruning criteria is sensitive in small models but not for large-scale models. It is a novel finding through this work. Overall, we demonstrate the superiority of the proposed model to the state-of-the-art models.

Building on the unprecedented capabilities of large language models for command understanding and zero-shot recognition of multi-modal vision-language transformers, visual language navigation (VLN) has emerged as an effective way to address multiple fundamental challenges toward a natural language interface to robot navigation. However, such vision-language models are inherently vulnerable due to the lack of semantic meaning of the underlying embedding space. Using a recently developed gradient based optimization procedure, we demonstrate that images can be modified imperceptibly to match the representation of totally different images and unrelated texts for a vision-language model. Building on this, we develop algorithms that can adversarially modify a minimal number of images so that the robot will follow a route of choice for commands that require a number of landmarks. We demonstrate that experimentally using a recently proposed VLN system; for a given navigation command, a robot can be made to follow drastically different routes. We also develop an efficient algorithm to detect such malicious modifications reliably based on the fact that the adversarially modified images have much higher sensitivity to added Gaussian noise than the original images.

Radiation therapy (RT) is one of the most effective treatments for cancer, and its success relies on the accurate delineation of targets. However, target delineation is a comprehensive medical decision that currently relies purely on manual processes by human experts. Manual delineation is time-consuming, laborious, and subject to interobserver variations. Although the advancements in artificial intelligence (AI) techniques have significantly enhanced the auto-contouring of normal tissues, accurate delineation of RT target volumes remains a challenge. In this study, we propose a visual language model-based RT target volume auto-delineation network termed Radformer. The Radformer utilizes a hierarichal vision transformer as the backbone and incorporates large language models to extract text-rich features from clinical data. We introduce a visual language attention module (VLAM) for integrating visual and linguistic features for language-aware visual encoding (LAVE). The Radformer has been evaluated on a dataset comprising 2985 patients with head-and-neck cancer who underwent RT. Metrics, including the Dice similarity coefficient (DSC), intersection over union (IOU), and 95th percentile Hausdorff distance (HD95), were used to evaluate the performance of the model quantitatively. Our results demonstrate that the Radformer has superior segmentation performance compared to other state-of-the-art models, validating its potential for adoption in RT practice.

Environmental Sound Classification is an important problem of sound recognition and is more complicated than speech recognition problems as environmental sounds are not well structured with respect to time and frequency. Researchers have used various CNN models to learn audio features from different audio features like log mel spectrograms, gammatone spectral coefficients, mel-frequency spectral coefficients, generated from the audio files, over the past years. In this paper, we propose a new methodology : Two-Level Classification; the Level 1 Classifier will be responsible to classify the audio signal into a broader class and the Level 2 Classifiers will be responsible to find the actual class to which the audio belongs, based on the output of the Level 1 Classifier. We have also shown the effects of different audio filters, among which a new method of Audio Crop is introduced in this paper, which gave the highest accuracies in most of the cases. We have used the ESC-50 dataset for our experiment and obtained a maximum accuracy of 78.75% in case of Level 1 Classification and 98.04% in case of Level 2 Classifications.

Recent advancements in large language model alignment leverage token-level supervisions to perform fine-grained preference optimization. However, existing token-level alignment methods either optimize on all available tokens, which can be noisy and inefficient, or perform selective training with complex and expensive key token selection strategies. In this work, we propose Selective Preference Optimization (SePO), a novel selective alignment strategy that centers on efficient key token selection. SePO proposes the first token selection method based on Direct Preference Optimization (DPO), which trains an oracle model to estimate a token-level reward function on the target data. This method applies to any existing alignment datasets with response-level annotations and enables cost-efficient token selection with small-scale oracle models and training data. The estimated reward function is then utilized to score all tokens within the target dataset, where only the key tokens are selected to supervise the target policy model with a reference model-free contrastive objective function. Extensive experiments on three public evaluation benchmarks show that SePO significantly outperforms competitive baseline methods by only optimizing 30% key tokens on the target dataset. SePO applications on weak-to-strong generalization show that weak oracle models effectively supervise strong policy models with up to 16.8x more parameters. SePO also effectively selects key tokens from out-of-distribution data to enhance strong policy models and alleviate the over-optimization problem.

Drug repurposing offers a promising avenue for accelerating drug development by identifying new therapeutic potentials of existing drugs. In this paper, we propose a multi-agent framework to enhance the drug repurposing process using state-of-the-art machine learning techniques and knowledge integration. Our framework comprises several specialized agents: an AI Agent trains robust drug-target interaction (DTI) models; a Knowledge Graph Agent utilizes the drug-gene interaction database (DGIdb), DrugBank, Comparative Toxicogenomics Database (CTD), and Search Tool for Interactions of Chemicals (STITCH) to systematically extract DTIs; and a Search Agent interacts with biomedical literature to annotate and verify computational predictions. By integrating outputs from these agents, our system effectively harnesses diverse data sources, including external databases, to propose viable repurposing candidates. Preliminary results demonstrate the potential of our approach in not only predicting drug-disease interactions but also in reducing the time and cost associated with traditional drug discovery methods. This paper highlights the scalability of multi-agent systems in biomedical research and their role in driving innovation in drug repurposing. Our approach not only outperforms existing methods in predicting drug repurposing potential but also provides interpretable results, paving the way for more efficient and cost-effective drug discovery processes.

This paper considers the problem of Multi-Hop Video Question Answering (MH-VidQA) in long-form egocentric videos. This task not only requires to answer visual questions, but also to localize multiple relevant time intervals within the video as visual evidences. We develop an automated pipeline to create multi-hop question-answering pairs with associated temporal evidence, enabling to construct a large-scale dataset for instruction-tuning. To monitor the progress of this new task, we further curate a high-quality benchmark, MultiHop-EgoQA, with careful manual verification and refinement. Experimental results reveal that existing multi-modal systems exhibit inadequate multi-hop grounding and reasoning abilities, resulting in unsatisfactory performance. We then propose a novel architecture, termed as Grounding Scattered Evidence with Large Language Model (GeLM), that enhances multi-modal large language models (MLLMs) by incorporating a grounding module to retrieve temporal evidence from videos using flexible grounding tokens. Trained on our visual instruction data, GeLM demonstrates improved multi-hop grounding and reasoning capabilities, setting a new baseline for this challenging task. Furthermore, when trained on third-person view videos, the same architecture also achieves state-of-the-art performance on the single-hop VidQA benchmark, ActivityNet-RTL, demonstrating its effectiveness.