金融/语音/音频处理学术速递[12.23]

q-fin金融，共计12篇

cs.SD语音，共计2篇

eess.AS音频处理，共计2篇

1.q-fin金融:

【1】 Optimal Portfolio Choice and Stock Centrality for Tail Risk Events 标题：尾部风险事件的最优投资组合选择与股票中心性 链接：https://arxiv.org/abs/2112.12031

作者：Christis Katsouris 机构：University of Southampton 摘要：我们提出了一个新的风险矩阵来描述投资者的最优投资组合选择的尾部关注。矩阵的对角线包含投资组合中每项资产的风险价值，非对角线包含反映资产之间尾部连接的成对Delta-CoVaR度量。首先，我们推导了相关的二次风险函数具有闭式解的条件。其次，我们检验了投资组合风险和特征向量中心性之间的关系。第三，我们证明了投资组合风险不一定随着股票中心性的增加而增加。第四，我们证明了在一定条件下，资产中心性增加了资产对投资组合的最优权重分配。总体而言，我们的实证研究表明，基于夏普比率检验的高连通性优于低连通性的网络拓扑。 摘要：We propose a novel risk matrix to characterize the optimal portfolio choice of an investor with tail concerns. The diagonal of the matrix contains the Value-at-Risk of each asset in the portfolio and the off-diagonal the pairwise Delta-CoVaR measures reflecting tail connections between assets. First, we derive the conditions under which the associated quadratic risk function has a closed-form solution. Second, we examine the relationship between portfolio risk and eigenvector centrality. Third, we show that portfolio risk is not necessarily increasing with respect to stock centrality. Forth, we demonstrate under certain conditions that asset centrality increases the optimal weight allocation of the asset to the portfolio. Overall, our empirical study indicates that a network topology which exhibits low connectivity is outperformed by high connectivity based on a Sharpe ratio test.

【2】 Spread Option Pricing in a Copula Affine GARCH(p,q) Model 标题：Copula仿射GARCH(p，q)模型中的价差期权定价 链接：https://arxiv.org/abs/2112.11968

作者：Edoardo Berton,Lorenzo Mercuri 机构：Department of Economics, Management and Quantitative Methods, University, of Milan., CREST Japan Science and Technology Agency. 摘要：在本研究中，我们将copula函数与仿射GARCH（p，q）过程相结合，构建了一个二元市场模型，用于描述原木价格的边际动态。然后，我们提供了一个欧洲价差期权合约定价的数值过程。为了评估我们方法的准确性，我们与蒙特卡罗模拟方法进行了比较。 摘要：In this study, we construct a bivariate market model combining the copula function with the affine GARCH(p,q) process used to describe the marginal dynamics of the log price. We then provide a numerical procedure for pricing European spread option contracts. To assess the accuracy of our approach we present a comparison with the Monte Carlo simulation method.

【3】 Principal agent mean field games in REC markets 标题：REC市场中的委托代理平均场博弈 链接：https://arxiv.org/abs/2112.11963

作者：Arvind Shrivats,Dena Firoozi,Sebastian Jaimungal 机构：edu)†Department of Decision Sciences 备注：Work in progress 摘要：委托代理博弈是一个日益增长的研究领域，其重点是委托人和代理人的最佳行为，前者从后者承包工作，以换取提供金钱奖励。虽然该领域规范地考虑单个代理，但多个代理，甚至无限多个代理由委托人签约的情况日益突出，并提出了有趣和现实的问题。在这里，代理人之间形成纳什均衡，作为集体的代理人与委托人之间形成斯塔克伯格均衡。我们将此框架应用于实施排放市场的问题。我们这样做的同时结合了市场清算和代理异质性，并通过结合MFG的概率方法而不是分析方法与现有文献区分开来，前者更自然地为我们的问题提供贷款。对于一个给定的市场设计，我们使用平均场博弈中的技术来发现代理之间的纳什均衡。然后，我们从监管者的角度提供优化市场设计的初步结果，监管者的目标是实现收入和总体环境效益的最大化。 摘要：Principal agent games are a growing area of research which focuses on the optimal behaviour of a principal and an agent, with the former contracting work from the latter, in return for providing a monetary award. While this field canonically considers a single agent, the situation where multiple agents, or even an infinite amount of agents are contracted by a principal are growing in prominence and pose interesting and realistic problems. Here, agents form a Nash equilibrium among themselves, and a Stackelberg equilibrium between themselves as a collective and the principal. We apply this framework to the problem of implementing emissions markets. We do so while incorporating market clearing as well as agent heterogeneity, and distinguish ourselves from extant literature by incorporating the probabilistic approach to MFGs as opposed to the analytic approach, with the former lending itself more naturally for our problem. For a given market design, we find the Nash equilibrium among agents using techniques from mean field games. We then provide preliminary results for the optimal market design from the perspective of the regulator, who aims to maximize revenue and overall environmental benefit.

【4】 Product traits, decision-makers, and household low-carbon technology adoptions: moving beyond single empirical studies 标题：产品特性、决策者和家庭低碳技术采用：超越单一的实证研究 链接：https://arxiv.org/abs/2112.11867

作者：Emily Schulte,Fabian Scheller,Wilmer Pasut,Thomas Bruckner 机构：Institute for Infrastructure and Resources Management (IIRM), Leipzig University, Energy Economics and System Analysis, Division of Sustainability, Department of Technology, Management and Economics, Technical University of Denmark (DTU) 备注：None 摘要：尽管单个实证研究提供了关于谁因什么原因采用特定LCT的重要见解，但有关决策主体（=谁决定）、决策对象（=决定什么）和背景（=何时何地决定）之间关系的基本问题仍然没有答案。在本文中，通过推导住宅决策的决策框架来解决这一研究缺口，表明决策主体和对象的特征是采用前财务、环境、符号、规范、努力和技术考虑的决定因素。此后，通过采用光伏系统、节能电器和绿色关税方面的文献对决策框架进行初步验证。在六项提议的关系中，两项可以确认（财务和环境），一项可以拒绝（努力），三项由于缺乏证据既不能确认也不能拒绝。未来关于LCT采用的研究可以使用决策框架作为指导，以建立更协调和综合的方法，最终解决基本问题。 摘要：Although single empirical studies provide important insights into who adopts a specific LCT for what reason, fundamental questions concerning the relations between decision subject (= who decides), decision object (= what is decided upon) and context (= when and where it is decided) remain unanswered. In this paper, this research gap is addressed by deriving a decision framework for residential decision-making, suggesting that traits of decision subject and object are determinants of financial, environmental, symbolic, normative, effort and technical considerations preceding adoption. Thereafter, the decision framework is initially verified by employing literature on the adoption of photovoltaic systems, energy-efficient appliances and green tariffs. Of the six proposed relations, two could be confirmed (financial and environmental), one could be rejected (effort), and three could neither be confirmed nor rejected due to lacking evidence. Future research on LCT adoption could use the decision framework as a guidepost to establish a more coordinated and integrated approach, ultimately allowing to address fundamental questions.

【5】 The co-evolutionary relationship between digitalization and organizational agility: Ongoing debates, theoretical developments and future research perspectives 标题：数字化与组织高速移动的协同进化关系：争论、理论发展与未来研究展望 链接：https://arxiv.org/abs/2112.11822

作者：Francesco Ciampi,Monica Faraoni,Jacopo Ballerini,Francesco Meli 备注：None 摘要：本研究首次对数字化与组织敏捷性（OA）之间关系的文献进行了系统回顾。它将书目耦合方法应用于2021年6月30日出版的171篇同行评议稿中。它使用数字化的观点来调查旨在为企业提供有效应对日益动荡的环境所需的灵活性的过程的促成因素、障碍和好处。发现并分析了三个不同但相互关联的主题集群，分别侧重于作为OA关键驱动因素的大数据分析能力、供应链层面的数字化和敏捷性之间的关系以及信息技术能力在改善OA中的作用。通过采用动态能力视角，本研究克服了传统观点，即主要考虑OA的数字能力使能因素，而不是可能的结果。我们的研究结果表明，数字化和OA之间的关系除了复杂之外，还具有双向性。本研究还确定了现存的研究差距，并就未来可能的研究途径和新的管理解决方案提出了13个原始研究命题。 摘要：This study is the first to provide a systematic review of the literature focused on the relationship between digitalization and organizational agility (OA). It applies the bibliographic coupling method to 171 peer-reviewed contributions published by 30 June 2021. It uses the digitalization perspective to investigate the enablers, barriers and benefits of processes aimed at providing firms with the agility required to effectively face increasingly turbulent environments. Three different, though interconnected, thematic clusters are discovered and analysed, respectively focusing on big-data analytic capabilities as crucial drivers of OA, the relationship between digitalization and agility at a supply chain level, and the role of information technology capabilities in improving OA. By adopting a dynamic capabilities perspective, this study overcomes the traditional view, which mainly considers digital capabilities enablers of OA, rather than as possible outcomes. Our findings reveal that, in addition to being complex, the relationship between digitalization and OA has a bidirectional character. This study also identifies extant research gaps and develops 13 original research propositions on possible future research pathways and new managerial solutions.

【6】 Double Standards: The Implications of Near Certainty Drone Strikes in Pakistan 标题：双重标准：巴基斯坦近乎确定的无人机袭击的含义 链接：https://arxiv.org/abs/2112.11565

作者：Shyam Raman,Paul Lushenko,Sarah Kreps 机构：Brooks School of Public Policy, Cornell University, Ithaca, NY 摘要：2013年，美国总统奥巴马宣布了一项政策，以尽量减少未申报战区无人驾驶飞机袭击后的平民伤亡。这项政策使奥巴马总统对罢工的批准与几乎肯定不会有平民伤亡的事实相矛盾。学者们没有实证研究奥巴马政策的优点。相反，他们依靠巴基斯坦平民伤亡的描述性趋势来证明政策影响的相互竞争的说法是正确的。我们对奥巴马政策对美国无人机袭击后巴基斯坦平民伤亡的影响提供了一个新的估计。我们采用回归不连续设计来评估奥巴马政策对平民伤亡、打击精度和公布的平民伤亡的影响。我们发现，在奥巴马的政策宣布前大约两年，平民伤亡出现了间断，这证实了我们的主要研究，包括采访负责实施近乎确定标准的高级官员。在确认大幅削减后，我们估计该政策每月减少12名平民死亡，或每次罢工减少2人伤亡。该政策还提高了美国无人机打击的精确度，使其只杀死预定目标。最后，我们使用蒙特卡罗模拟来估计该政策涉及320名平民伤亡。然后，我们进行了统计生命值计算，以显示广告中的平民伤亡代表了8000万至2.6亿美元的收益。因此，除了制约社会和政治结果外，近乎确定的标准还带来了经济影响，而这方面的研究要少得多。 摘要：In 2013, U.S. President Barack Obama announced a policy to minimize civilian casualties following drone strikes in undeclared theaters of war. The policy calibrated Obamas approval of strikes against the near certainty of no civilian casualties. Scholars do not empirically study the merits of Obamas policy. Rather, they rely on descriptive trends for civilian casualties in Pakistan to justify competing claims for the policys impact. We provide a novel estimate for the impact of Obamas policy for civilian casualties in Pakistan following U.S. drone strikes. We employ a regression discontinuity design to estimate the effect of Obamas policy for civilian casualties, strike precision, and adverted civilian casualties. We find a discontinuity in civilian casualties approximately two years before Obamas policy announcement, corroborating our primary research including interviews with senior officials responsible for implementing the near certainty standard. After confirming the sharp cutoff, we estimate the policy resulted in a reduction of 12 civilian deaths per month or 2 casualties per strike. The policy also enhanced the precision of U.S. drone strikes to the point that they only killed the intended targets. Finally, we use a Monte Carlo simulation to estimate that the policy adverted 320 civilian casualties. We then conduct a Value of Statistical Life calculation to show that the adverted civilian casualties represent a gain of 80 to 260 million U.S. dollars. In addition to conditioning social and political outcomes, then, the near certainty standard also imposed economic implications that are much less studied.

【7】 Associational and plausible causal effects of COVID-19 public health policies on economic and mental distress 标题：冠状病毒公共卫生政策对经济和精神困扰的关联和似是而非的因果效应 链接：https://arxiv.org/abs/2112.11564

作者：Reka Sundaram-Stukel,Richard J Davidson 机构：Summary, Background, The COVID-, pandemic has increased mental distress globally. The proportion of people reporting, anxiety is ,%, and depression is ,% points. Disentangling associational and causal contributions of 备注：Pages 8, figures 2 in main text 摘要：背景2019冠状病毒疾病流行在全球范围内增加了精神痛苦。报告焦虑的比例为26%，抑郁的比例为34%。排除行为和CO2019冠状病毒疾病的因果关联和经济困境对心理困扰的影响将决定不同的缓解策略，以减少长期与大流行相关的精神痛苦。方法采用家庭脉搏调查（HPS）2020年4月至2021年2月的数据，调查美国公民的精神痛苦归因于COVID-19。我们将HPS 2019冠状病毒疾病调查数据与国家级每周：COVID-19病例和疾病控制中心、公共政策和苹果和谷歌的流动性数据的死亡数据相结合。最后，我们构建了经济和心理困扰测度，用滞后因变量估计结构模型，梳理出公共卫生政策对经济和心理困扰的关联和因果路径系数。从2020年4月到2021年2月的调查结果中，我们发现，美国的焦虑和抑郁在设计上稳步上升，流动性限制主要影响了公共卫生政策，其中企业和餐馆吸收了最大的冲击。T-2019冠状病毒疾病的发病率增加了4.1%，经济损失增加了6.3%个百分点。t-1期的失业和住房不安全感分别增加了t期精神困扰的29.1%和32.7%。然而，t-1期食物不安全感在t期内使精神痛苦降低了4.9%。t-1期经济痛苦对t期抑郁症的流行相关潜在因果路径系数为57.8%，焦虑为55.9%。因此，我们2019冠状病毒疾病的情况下，行为和经济窘迫可能与流行相关的时期T精神痛苦因果关系。 摘要：Background The COVID-19 pandemic has increased mental distress globally. The proportion of people reporting anxiety is 26%, and depression is 34% points. Disentangling associational and causal contributions of behavior, COVID-19 cases, and economic distress on mental distress will dictate different mitigation strategies to reduce long-term pandemic-related mental distress. Methods We use the Household Pulse Survey (HPS) April 2020 to February 2021 data to examine mental distress among U.S. citizens attributable to COVID-19. We combined HPS survey data with publicly available state-level weekly: COVID-19 case and death data from the Centers for Disease Control, public policies, and Apple and Google mobility data. Finally, we constructed economic and mental distress measures to estimate structural models with lag dependent variables to tease out public health policies' associational and causal path coefficients on economic and mental distress. Findings From April 2020 to February 2021, we found that anxiety and depression had steadily climbed in the U.S. By design, mobility restrictions primarily affected public health policies where businesses and restaurants absorbed the biggest hit. Period t-1 COVID-19 cases increased job loss by 4.1% and economic distress by 6.3% points in the same period. Job-loss and housing insecurity in t-1 increased period t mental distress by 29.1% and 32.7%, respectively. However, t-1 food insecurity decreased mental distress by 4.9% in time t. The pandemic-related potential causal path coefficient of period t-1 economic distress on period t depression is 57.8%, and anxiety is 55.9%. Thus, we show that period t-1 COVID-19 case information, behavior, and economic distress may be causally associated with pandemic related period t mental distress.

【8】 Cultural Diversity and Its Impact on Governance 标题：文化多样性及其对治理的影响 链接：https://arxiv.org/abs/2112.11563

作者：Tomáš Evan,Vladimír Holý 机构：Czech Technical University in Prague, Thákurova ,, Prague , Czechia, Prague University of Economics and Business, Winston Churchill Square , Prague , Czechia 摘要：霍夫斯泰德的六个文化维度使衡量国家文化成为可能，但因假设每个国家的同质性而受到批评。在本文中，我们基于霍夫斯泰德的文化维度提出了两项措施，其中考虑到公民相对于其原籍国的异质结构。通过这些改进措施，我们研究了异质文化和文化多样性对由六个全球治理指标衡量的机构质量的影响。我们使用线性回归模型，考虑到空间和时间维度的依赖性以及治理指标之间的高度相关性。我们的研究结果表明，文化多样性的影响改善了一些治理指标，而恶化了其他指标，这取决于霍夫斯泰德个人的文化维度。 摘要：Hofstede's six cultural dimensions make it possible to measure the culture of countries but are criticized for assuming the homogeneity of each country. In this paper, we propose two measures based on Hofstede's cultural dimensions which take into account the heterogeneous structure of citizens with respect to their countries of origin. Using these improved measures, we study the influence of heterogeneous culture and cultural diversity on the quality of institutions measured by the six worldwide governance indicators. We use a linear regression model allowing for dependence in spatial and temporal dimensions as well as high correlation between the governance indicators. Our results show that the effect of cultural diversity improves some of the governance indicators while worsening others depending on the individual Hofstede cultural dimension.

【9】 Can large-scale R&I funding stimulate post-crisis recovery growth? Evidence for Finland during COVID-19 标题：大规模的R&I资金能刺激危机后的复苏增长吗？芬兰在冠状病毒期间的证据 链接：https://arxiv.org/abs/2112.11562

作者：Timo Mitze,Teemu Makkonen 机构： Timo Mitze 1 and Teemu Makkonen 2 1 University of Southern Denmark, org0000‐000 3‐ 3799‐ 5 200 2 University of Eastern Finland 备注：30 pages, 6 figures, 5 tables 摘要：2019冠状病毒疾病流行和随后的公共卫生限制导致全球经济活动大幅滑坡。这一衰退受到各种政策行动的影响，以缓解CVID-19危机的不利社会经济后果，并最终使经济重回正轨。我们对芬兰大量增加研究与创新（R&I）资金的有效性进行了事前评估，以通过增加芬兰企业的研究与创新活动来刺激危机后复苏增长。我们利用了这样一个事实，即2020年授予处于破坏性环境中的公司的新型R&I补助金是通过既定的R&I政策渠道分配的。这使我们能够利用新冠肺炎前的数据估计芬兰坚果-3地区研发资金与经济增长之间的结构性联系。然后，利用估计值对样本外的区域复苏增长进行预测，并量化R&I资金的增长贡献。根据所选择的情况，我们的预测指出，在2020，GDP的平均回收率增长率在2021至2.5%之间，下降幅度达到了-2%。研发资金构成了复苏过程的重要支柱，对GDP增长的平均贡献在0.4%到1%之间。 摘要：The COVID-19 pandemic and subsequent public health restrictions led to a significant slump in economic activities around the globe. This slump has met by various policy actions to cushion the detrimental socio-economic consequences of the COVID-19 crisis and eventually bring the economy back on track. We provide an ex-ante evaluation of the effectiveness of a massive increase in research and innovation (R&I) funding in Finland to stimulate post-crisis recovery growth through an increase in R&I activities of Finnish firms. We make use of the fact that novel R&I grants for firms in disruptive circumstances granted in 2020 were allocated through established R&I policy channels. This allows us to estimate the structural link between R&I funding and economic growth for Finnish NUTS-3 regions using pre-COVID-19 data. Estimates are then used to forecast regional recovery growth out of sample and to quantify the growth contribution of R&I funding. Depending on the chosen scenario, our forecasts point to a mean recovery growth rate of GDP between 2-4% in 2021 after a decline of up to -2.5% in 2020. R&I funding constitutes a significant pillar of the recovery process with mean contributions in terms of GDP growth of between 0.4% and 1%.

【10】 The Changing Role of Entrepreneurial Universities in the Altering Innovation Policy: Opportunities Arising from the Paradigm Change in Light of the Experience of Széchenyi István University 标题：创业型大学在变革创新政策中的角色转变：从Széchenyi István大学的经验看范式变革带来的机遇 链接：https://arxiv.org/abs/2112.11499

作者：Attila Lajos Makai,Szabolcs Rámháp 机构：The Changing Role of, Entrepreneurial Universities in, the Altering Innovation Policy:, Opportunities Arising from the, Paradigm Change in Light of, the Experience of Széchenyi, Summary 备注：None 摘要：创业型大学是匈牙利高等教育模式变革后的一个新兴类别，它所取得的进步不仅加深了大学与产业之间的关系，也强化了技术和知识转移过程，但也增加了大学在制定区域创新政策中的作用。这一转变将区域创新生态系统参与者之间的合作以及经济、政府和学术系统之间的关系纳入一个新的框架。本文的目的是通过一个具体的例子来描述模型的变化过程，并概述大学参与当前不断变化的匈牙利创新政策体系的未来可能性。 摘要：The progress made by the entrepreneurial university, which is a newly emerging category in Hungarian higher education after its change of model, has not only deepened relations between universities and the industry and intensified the technology and knowledge transfer processes, but also increased the role of universities in shaping regional innovation policy. This transformation places co-operation between the actors of the regional innovation ecosystem and the relationships between the economic, governmental and academic systems into a new framework. The purpose of this paper is to describe the process of the change in the model through a specific example, and to outline the future possibilities of university involvement in the currently changing Hungarian innovation policy system.

【11】 Startup Ecosystem Rankings 标题：创业生态系统排名 链接：https://arxiv.org/abs/2112.11931

作者：Attila Lajos Makai 机构：Startup ecosystem rankings, PhD Student, Széchenyi István University, Hungary 备注：None 摘要：全球范围内，衡量创新绩效以及提供其空间框架的生态系统的实力和资源的排名数量、重要性和受欢迎程度呈上升趋势。除了影响经济参与者做出的具体决策外，这些排名还显著影响区域、国家和国际层面创新相关政策的制定。科学兴趣的增长证明了创业生态系统的重要性，越来越多的相关科学文章证明了这一点。创业生态系统的概念是一个相对较新的范畴，自本世纪末以来，它在日常生活和科学生活中的应用日益广泛。当然，与此同时，决策者和学者也对可测量性和可比性提出了要求。这一需求通过创业生态系统排名得以满足，该排名现在在大陆和全球范围内衡量和排名各个生态系统的表现。然而，尽管研究高等教育、经济表现甚至创新排名的科学出版物数量可以用数千份来衡量，但迄今为止，科学研究很少或根本没有涉及创业生态系统的排名。本研究和相关研究旨在通过介绍和分析全球排名的特点，并确定未来可能的研究方向，填补这一空白。 摘要：The number, importance, and popularity of rankings measuring innovation performance and the strength and resources of ecosystems that provide its spatial framework are on an increasing trend globally. In addition to influencing the specific decisions taken by economic actors, these rankings significantly impact the development of innovation-related policies at regional, national, and international levels. The importance of startup ecosystems is proven by the growing scientific interest, which is demonstrated by the increasing number of related scientific articles. The concept of the startup ecosystem is a relatively new category, the application of which in everyday and scientific life has been gaining ground since the end of the 2000s. In parallel, of course, the demand for measurability and comparability has emerged among decision-makers and scholars. This demand is met by startup ecosystem rankings, which now measure and rank the performance of individual ecosystems on a continental and global scale. However, while the number of scientific publications examining rankings related to higher education, economic performance, or even innovation, can be measured in the order of thousands, scientific research has so far rarely or tangentially addressed the rankings of startup ecosystems. This study and the related research intend to fill this gap by presenting and analysing the characteristics of global rankings and identifying possible future research directions.

【12】 Mild to classical solutions for XVA equations under stochastic volatility 标题：随机波动条件下XVA方程的温和到经典解 链接：https://arxiv.org/abs/2112.11808

作者：Damiano Brigo,Federico Graceffa,Alexander Kalinin 摘要：我们将存在违约、抵押品和融资的或有债权的估值扩展到随机函数设置，并通过鞅描述违约前的价值过程。预设值前半鞅也可以用随机路径相关系数和鞅作为驱动的BSDE来描述。在此过程中，我们通过放宽可用市场信息的条件，允许任意无违约过滤，并构建广泛的违约时间类别，来概括以前的设置。此外，在随机波动下，我们通过抛物型半线性偏微分方程的温和解刻画了违约前的价值过程，并给出了温和解唯一存在和经典的充分条件。 摘要：We extend the valuation of contingent claims in presence of default, collateral and funding to a random functional setting and characterise pre-default value processes by martingales. Pre-default value semimartingales can also be described by BSDEs with random path-dependent coefficients and martingales as drivers. En route, we generalise previous settings by relaxing conditions on the available market information, allowing for an arbitrary default-free filtration and constructing a broad class of default times. Moreover, under stochastic volatility, we characterise pre-default value processes via mild solutions to parabolic semilinear PDEs and give sufficient conditions for mild solutions to exist uniquely and to be classical.

2.cs.SD语音:

【1】 Mixed Precision of Quantization of Transformer Language Models for Speech Recognition 标题：用于语音识别的Transformer语言模型量化的混合精度 链接：https://arxiv.org/abs/2112.11540

作者：Junhao Xu,Shoukang Hu,Jianwei Yu,Xunying Liu,Helen Meng 机构：The Chinese University of Hong Kong, Hong Kong SAR, China 备注：arXiv admin note: substantial text overlap with arXiv:2112.11438, arXiv:2111.14479 摘要：以Transformer为代表的最先进的神经语言模型在实际应用中变得越来越复杂和昂贵。低比特深度神经网络量化技术提供了一个强大的解决方案，以大大减少其模型的大小。当前的低比特量化方法基于统一的精度，无法考虑系统不同部分对量化误差的不同性能灵敏度。为此，本文提出了一种新的混合精度DNN量化方法。使用两种技术自动学习最佳局部精度设置。第一种是基于Hessian迹加权量化扰动形式的量化灵敏度度量。第二种是基于混合精度的Transformer结构搜索。采用交替方向乘子法（ADMM）有效地训练混合精度量化DNN系统。在Penn Treebank（PTB）和交换机语料库训练的LF-MMI TDNN系统上进行的实验表明，所提出的混合精度Transformer量化技术在不降低识别性能的情况下，在全精度基线上实现了高达16倍的模型大小压缩比。当用于压缩具有更多层的更大全精度TransformerLM时，可获得高达1.7%绝对（18%相对）的总体字错误率（WER）降低。 摘要：State-of-the-art neural language models represented by Transformers are becoming increasingly complex and expensive for practical applications. Low-bit deep neural network quantization techniques provides a powerful solution to dramatically reduce their model size. Current low-bit quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of the system to quantization errors. To this end, novel mixed precision DNN quantization methods are proposed in this paper. The optimal local precision settings are automatically learned using two techniques. The first is based on a quantization sensitivity metric in the form of Hessian trace weighted quantization perturbation. The second is based on mixed precision Transformer architecture search. Alternating direction methods of multipliers (ADMM) are used to efficiently train mixed precision quantized DNN systems. Experiments conducted on Penn Treebank (PTB) and a Switchboard corpus trained LF-MMI TDNN system suggest the proposed mixed precision Transformer quantization techniques achieved model size compression ratios of up to 16 times over the full precision baseline with no recognition performance degradation. When being used to compress a larger full precision Transformer LM with more layers, overall word error rate (WER) reductions up to 1.7% absolute (18% relative) were obtained.

【2】 Self-Supervised Learning based Monaural Speech Enhancement with Multi-Task Pre-Training 标题：基于自监督学习的多任务预训练单声道语音增强 链接：https://arxiv.org/abs/2112.11459

作者：Yi Li,Yang Sun,Syed Mohsen Naqvi 机构： Intelligent Sensing and Communications Research Group, Newcastle University, UK, Big Data Institute, University of Oxford, UK 备注：Submitted to ICASSP 2022. arXiv admin note: text overlap with arXiv:2112.11142 摘要：在自监督学习中，利用已有的预任务来减小估计语音信号和目标语音信号的增强性能之间的差距是一个挑战。在本文中，我们提出了一种多任务预训练方法，通过自监督学习来提高语音增强性能。在预训练自动编码器（PAE）中，只需要一组有限的干净语音信号来学习它们的潜在表示。同时，为了解决单个预任务的局限性，提出的掩蔽模块利用去冗余掩蔽和估计比率掩蔽对混合信号进行去噪，作为第二个预任务。与PAE不同，PAE对目标语音信号进行估计，下游任务自动编码器（DAE）利用大量未标记和看不见的混响混合来生成估计的混合。经过训练的DAE由学习的表示和掩码共享。在一个基准数据集上的实验结果表明，该方法优于现有的方法。 摘要：In self-supervised learning, it is challenging to reduce the gap between the enhancement performance on the estimated and target speech signals with existed pre-tasks. In this paper, we propose a multi-task pre-training method to improve the speech enhancement performance with self-supervised learning. Within the pre-training autoencoder (PAE), only a limited set of clean speech signals are required to learn their latent representations. Meanwhile, to solve the limitation of single pre-task, the proposed masking module exploits the dereverberated mask and estimated ratio mask to denoise the mixture as the second pre-task. Different from the PAE, where the target speech signals are estimated, the downstream task autoencoder (DAE) utilizes a large number of unlabeled and unseen reverberant mixtures to generate the estimated mixtures. The trained DAE is shared by the learned representations and masks. Experimental results on a benchmark dataset demonstrate that the proposed method outperforms the state-of-the-art approaches.

3.eess.AS音频处理:

【1】 The Phonetic Footprint of Parkinson's Disease 标题：帕金森病的语音足迹 链接：https://arxiv.org/abs/2112.11514

作者：Philipp Klumpp,Tomás Arias-Vergara,Juan Camilo Vásquez-Correa,Paula Andrea Pérez-Toro,Juan Rafael Orozco-Arroyave,Anton Batliner,Elmar Nöth 机构：Friedrich-Alexander-University Erlangen-Nuremberg, Pattern Recognition Lab, Martensstrasse , Erlangen, Germany, Universidad de Antioquia, Medellín, Colombia, Chair of Embedded Intelligence for Health Care and Wellbeing, University of Augsburg 备注：None 摘要：帕金森病（Parkinson'sdisease，PD）是最常见的神经退行性疾病之一，对患者的精细运动技能有重要影响。在言语产生和实现所需肌肉张力的过程中，不同发音者之间复杂的相互作用变得越来越困难，从而导致发音障碍。在受影响的个体中，经常可以观察到元音不稳定、发音模糊和说话缓慢等特征模式，并在以前的研究中进行分析，以确定PD的存在和进展。在这项工作中，我们使用专门针对健康语音数据训练的语音识别器来研究PD如何影响患者的语音足迹。尽管我们的系统以前从未见过任何病理性语言，但我们重新发现了在以前的贡献中描述的许多模式。此外，我们可以证明来自神经网络的中间激活可以作为编码个体疾病状态相关信息的特征向量。我们还能够直接将专家评定的说话人的可懂度与语音预测的平均置信度联系起来。我们的结果支持这样的假设，即训练能够分析PD语音的系统不一定需要病理数据。 摘要：As one of the most prevalent neurodegenerative disorders, Parkinson's disease (PD) has a significant impact on the fine motor skills of patients. The complex interplay of different articulators during speech production and realization of required muscle tension become increasingly difficult, thus leading to a dysarthric speech. Characteristic patterns such as vowel instability, slurred pronunciation and slow speech can often be observed in the affected individuals and were analyzed in previous studies to determine the presence and progression of PD. In this work, we used a phonetic recognizer trained exclusively on healthy speech data to investigate how PD affected the phonetic footprint of patients. We rediscovered numerous patterns that had been described in previous contributions although our system had never seen any pathological speech previously. Furthermore, we could show that intermediate activations from the neural network could serve as feature vectors encoding information related to the disease state of individuals. We were also able to directly correlate the expert-rated intelligibility of a speaker with the mean confidence of phonetic predictions. Our results support the assumption that pathological data is not necessarily required to train systems that are capable of analyzing PD speech.

【2】 Mixed Precision of Quantization of Transformer Language Models for Speech Recognition 标题：用于语音识别的Transformer语言模型量化的混合精度 链接：https://arxiv.org/abs/2112.11540

作者：Junhao Xu,Shoukang Hu,Jianwei Yu,Xunying Liu,Helen Meng 机构：The Chinese University of Hong Kong, Hong Kong SAR, China 备注：arXiv admin note: substantial text overlap with arXiv:2112.11438, arXiv:2111.14479 摘要：以Transformer为代表的最先进的神经语言模型在实际应用中变得越来越复杂和昂贵。低比特深度神经网络量化技术提供了一个强大的解决方案，以大大减少其模型的大小。当前的低比特量化方法基于统一的精度，无法考虑系统不同部分对量化误差的不同性能灵敏度。为此，本文提出了一种新的混合精度DNN量化方法。使用两种技术自动学习最佳局部精度设置。第一种是基于Hessian迹加权量化扰动形式的量化灵敏度度量。第二种是基于混合精度的Transformer结构搜索。采用交替方向乘子法（ADMM）有效地训练混合精度量化DNN系统。在Penn Treebank（PTB）和交换机语料库训练的LF-MMI TDNN系统上进行的实验表明，所提出的混合精度Transformer量化技术在不降低识别性能的情况下，在全精度基线上实现了高达16倍的模型大小压缩比。当用于压缩具有更多层的更大全精度TransformerLM时，可获得高达1.7%绝对（18%相对）的总体字错误率（WER）降低。 摘要：State-of-the-art neural language models represented by Transformers are becoming increasingly complex and expensive for practical applications. Low-bit deep neural network quantization techniques provides a powerful solution to dramatically reduce their model size. Current low-bit quantization methods are based on uniform precision and fail to account for the varying performance sensitivity at different parts of the system to quantization errors. To this end, novel mixed precision DNN quantization methods are proposed in this paper. The optimal local precision settings are automatically learned using two techniques. The first is based on a quantization sensitivity metric in the form of Hessian trace weighted quantization perturbation. The second is based on mixed precision Transformer architecture search. Alternating direction methods of multipliers (ADMM) are used to efficiently train mixed precision quantized DNN systems. Experiments conducted on Penn Treebank (PTB) and a Switchboard corpus trained LF-MMI TDNN system suggest the proposed mixed precision Transformer quantization techniques achieved model size compression ratios of up to 16 times over the full precision baseline with no recognition performance degradation. When being used to compress a larger full precision Transformer LM with more layers, overall word error rate (WER) reductions up to 1.7% absolute (18% relative) were obtained.

机器翻译，仅供参考