The identification of sensory cues associated with potential opportunities and dangers is frequently complicated by unrelated events that separate useful cues by long delays. As a result, it remains a challenging task for state-of-the-art spiking neural networks (SNNs) to identify long-term temporal dependencies since bridging the temporal gap necessitates an extended memory capacity. To address this challenge, we propose a novel biologically inspired Long Short-Term Memory Leaky Integrate-and-Fire spiking neuron model, dubbed LSTM-LIF. Our model incorporates carefully designed somatic and dendritic compartments that are tailored to retain short- and long-term memories. The theoretical analysis further confirms its effectiveness in addressing the notorious vanishing gradient problem. Our experimental results, on a diverse range of temporal classification tasks, demonstrate superior temporal classification capability, rapid training convergence, strong network generalizability, and high energy efficiency of the proposed LSTM-LIF model. This work, therefore, opens up a myriad of opportunities for resolving challenging temporal processing tasks on emerging neuromorphic computing machines.
The goal of Automatic Voice Over (AVO) is to generate speech in sync with a silent video given its text script. Recent AVO frameworks built upon text-to-speech synthesis (TTS) have shown impressive results. However, the current AVO learning objective of acoustic feature reconstruction brings in indirect supervision for inter-modal alignment learning, thus limiting the synchronization performance and synthetic speech quality. To this end, we propose a novel AVO method leveraging the learning objective of self-supervised discrete speech unit prediction, which not only provides more direct supervision for the alignment learning, but also alleviates the mismatch between the text-video context and acoustic features. Experimental results show that our proposed method achieves remarkable lip-speech synchronization and high speech quality by outperforming baselines in both objective and subjective evaluations. Code and speech samples are publicly available.
In active speaker detection (ASD), we would like to detect whether an on-screen person is speaking based on audio-visual cues. Previous studies have primarily focused on modeling audio-visual synchronization cue, which depends on the video quality of the lip region of a speaker. In real-world applications, it is possible that we can also have the reference speech of the on-screen speaker. To benefit from both facial cue and reference speech, we propose the Target Speaker TalkNet (TS-TalkNet), which leverages a pre-enrolled speaker embedding to complement the audio-visual synchronization cue in detecting whether the target speaker is speaking. Our framework outperforms the popular model, TalkNet on two datasets, achieving absolute improvements of 1.6\% in mAP on the AVA-ActiveSpeaker validation set, and 0.8\%, 0.4\%, and 0.8\% in terms of AP, AUC and EER on the ASW test set, respectively. Code is available at \href{https://github.com/Jiang-Yidi/TS-TalkNet/}{\color{red}{https://github.com/Jiang-Yidi/TS-TalkNet/}}.
The biological neural systems evolved to adapt to ecological environment for efficiency and effectiveness, wherein neurons with heterogeneous structures and rich dynamics are optimized to accomplish complex cognitive tasks. Most of the current research of biologically inspired spiking neural networks (SNNs) are, however, grounded on a homogeneous neural coding scheme, which limits their overall performance in terms of accuracy, latency, efficiency, and robustness, etc. In this work, we argue that one should holistically design the network architecture to incorporate diverse neuronal functions and neural coding schemes for best performance. As an early attempt in this research direction, we put forward a hybrid neural coding framework that integrates multiple neural coding schemes discovered in neuroscience. We demonstrate that the proposed hybrid coding scheme achieves a comparable accuracy with the state-of-the-art SNNs with homogeneous neural coding on CIFAR-10, CIFAR-100, and Tiny-ImageNet datasets with less than eight time steps and at least 3.90x fewer computations. Furthermore, we demonstrate accurate, rapid, and robust sound source localization on SoClas dataset. This study yields valuable insights into the performance of various hybrid neural coding designs and hold significant implications for designing high performance SNNs.
Audio Deepfake Detection (ADD) aims to detect the fake audio generated by text-to-speech (TTS), voice conversion (VC) and replay, etc., which is an emerging topic. Traditionally we take the mono signal as input and focus on robust feature extraction and effective classifier design. However, the dual-channel stereo information in the audio signal also includes important cues for deepfake, which has not been studied in the prior work. In this paper, we propose a novel ADD model, termed as M2S-ADD, that attempts to discover audio authenticity cues during the mono-to-stereo conversion process. We first projects the mono to a stereo signal using a pretrained stereo synthesizer, then employs a dual-branch neural architecture to process the left and right channel signals, respectively. In this way, we effectively reveal the artifacts in the fake audio, thus improve the ADD performance. The experiments on the ASVspoof2019 database show that M2S-ADD outperforms all baselines that input mono. We release the source code at \url{https://github.com/AI-S2-Lab/M2S-ADD}.
In this paper, we present HuatuoGPT, a large language model (LLM) for medical consultation. The core recipe of HuatuoGPT is to leverage both \textit{distilled data from ChatGPT} and \textit{real-world data from doctors} in the supervised fine-tuned stage. The responses of ChatGPT are usually detailed, well-presented and informative while it cannot perform like a doctor in many aspects, e.g. for integrative diagnosis. We argue that real-world data from doctors would be complementary to distilled data in the sense the former could tame a distilled language model to perform like doctors. To better leverage the strengths of both data, we train a reward model to align the language model with the merits that both data bring, following an RLAIF (reinforced learning from AI feedback) fashion. To evaluate and benchmark the models, we propose a comprehensive evaluation scheme (including automatic and manual metrics). Experimental results demonstrate that HuatuoGPT achieves state-of-the-art results in performing medical consultation among open-source LLMs in GPT-4 evaluation, human evaluation, and medical benchmark datasets. It is worth noting that by using additional real-world data and RLAIF, the distilled language model (i.e., HuatuoGPT) outperforms its teacher model ChatGPT in most cases. Our code, data, and models are publicly available at \url{https://github.com/FreedomIntelligence/HuatuoGPT}. The online demo is available at \url{https://www.HuatuoGPT.cn/}.
Topic segmentation and outline generation strive to divide a document into coherent topic sections and generate corresponding subheadings. Such a process unveils the discourse topic structure of a document that benefits quickly grasping and understanding the overall context of the document from a higher level. However, research and applications in this field have been restrained due to the lack of proper paragraph-level topic representations and large-scale, high-quality corpora in Chinese compared to the success achieved in English. Addressing these issues, we introduce a hierarchical paragraph-level topic structure representation with title, subheading, and paragraph that comprehensively models the document discourse topic structure. In addition, we ensure a more holistic representation of topic distribution within the document by using sentences instead of keywords to represent sub-topics. Following this representation, we construct the largest Chinese Paragraph-level Topic Structure corpus (CPTS), four times larger than the previously largest one. We also employ a two-stage man-machine collaborative annotation method to ensure the high quality of the corpus both in form and semantics. Finally, we validate the computability of CPTS on two fundamental tasks (topic segmentation and outline generation) by several strong baselines, and its efficacy has been preliminarily confirmed on the downstream task: discourse parsing. The representation, corpus, and benchmark we established will provide a solid foundation for future studies.
Training or finetuning large-scale language models (LLMs) such as GPT-3 requires substantial computation resources, motivating recent efforts to explore parameter-efficient adaptation to downstream tasks. One practical area of research is to treat these models as black boxes and interact with them through their inference APIs. In this paper, we investigate how to optimize few-shot text classification without accessing the gradients of the LLMs. To achieve this, we treat the black-box model as a feature extractor and train a classifier with the augmented text data. Data augmentation is performed using prompt-based finetuning on an auxiliary language model with a much smaller parameter size than the black-box model. Through extensive experiments on eight text classification datasets, we show that our approach, dubbed BT-Classifier, significantly outperforms state-of-the-art black-box few-shot learners and performs on par with methods that rely on full-model tuning.
Audio deepfake detection is an emerging topic in the artificial intelligence community. The second Audio Deepfake Detection Challenge (ADD 2023) aims to spur researchers around the world to build new innovative technologies that can further accelerate and foster research on detecting and analyzing deepfake speech utterances. Different from previous challenges (e.g. ADD 2022), ADD 2023 focuses on surpassing the constraints of binary real/fake classification, and actually localizing the manipulated intervals in a partially fake speech as well as pinpointing the source responsible for generating any fake audio. Furthermore, ADD 2023 includes more rounds of evaluation for the fake audio game sub-challenge. The ADD 2023 challenge includes three subchallenges: audio fake game (FG), manipulation region location (RL) and deepfake algorithm recognition (AR). This paper describes the datasets, evaluation metrics, and protocols. Some findings are also reported in audio deepfake detection tasks.
Discourse parsing, the task of analyzing the internal rhetorical structure of texts, is a challenging problem in natural language processing. Despite the recent advances in neural models, the lack of large-scale, high-quality corpora for training remains a major obstacle. Recent studies have attempted to overcome this limitation by using distant supervision, which utilizes results from other NLP tasks (e.g., sentiment polarity, attention matrix, and segmentation probability) to parse discourse trees. However, these methods do not take into account the differences between in-domain and out-of-domain tasks, resulting in lower performance and inability to leverage the high-quality in-domain data for further improvement. To address these issues, we propose a distant supervision framework that leverages the relations between topic structure and rhetorical structure. Specifically, we propose two distantly supervised methods, based on transfer learning and the teacher-student model, that narrow the gap between in-domain and out-of-domain tasks through label mapping and oracle annotation. Experimental results on the MCDTB and RST-DT datasets show that our methods achieve the best performance in both distant-supervised and supervised scenarios.