DER is the primary metric to evaluate diarization performance while facing a dilemma: the errors in short utterances or segments tend to be overwhelmed by longer ones. Short segments, e.g., `yes' or `no,' still have semantic information. Besides, DER overlooks errors in less-talked speakers. Although JER balances speaker errors, it still suffers from the same dilemma. Considering all those aspects, duration error, segment error, and speaker-weighted error constituting a complete diarization evaluation, we propose a Balanced Error Rate (BER) to evaluate speaker diarization. First, we propose a segment-level error rate (SER) via connected sub-graphs and adaptive IoU threshold to get accurate segment matching. Second, to evaluate diarization in a unified way, we adopt a speaker-specific harmonic mean between duration and segment, followed by a speaker-weighted average. Third, we analyze our metric via the modularized system, EEND, and the multi-modal method on real datasets. SER and BER are publicly available at https://github.com/X-LANCE/BER.

对话场景是语音处理技术最重要，最具挑战性的场景之一，因为对话中的人们以随意的方式相互反应。在对话中检测每个人的语音活动对于下游任务，例如自然语言处理，机器翻译等。人们指的是“何时说话”作为说话者诊断（SD）的检测技术。传统上，诊断错误率（DER）长期以来一直用作SD系统的标准评估度量。但是，der没有给简短的对话短语提供足够的重视，这在语义层面上很重要。此外，在语音社区中，仍然无法使用精心准确的手动测试数据集，适合评估对话性SD技术。在本文中，我们设计和描述了对话式短语扬声器诊断（CSSD）任务，该任务包括培训和测试数据集，评估指标和基线。在数据集方面，尽管先前开源的180小时对话魔术Data-RAMC数据集，但我们还准备了一个20小时的对话演讲测试数据集，并精心验证了CSSD任务的时间戳注释。在度量方面，我们设计了新的对话der（CDER）评估度量，该评估度量计算出语音级别的SD准确性。在基线方面，我们采用了一种常用的方法：变异贝叶斯HMM X-vector系统，作为CSSD任务的基线。我们的评估指标可在https://github.com/speechclub/cder_metric上公开获得。

Speaker embedding extractors significantly influence the performance of clustering-based speaker diarisation systems. Conventionally, only one embedding is extracted from each speech segment. However, because of the sliding window approach, a segment easily includes two or more speakers owing to speaker change points. This study proposes a novel embedding extractor architecture, referred to as a high-resolution embedding extractor (HEE), which extracts multiple high-resolution embeddings from each speech segment. Hee consists of a feature-map extractor and an enhancer, where the enhancer with the self-attention mechanism is the key to success. The enhancer of HEE replaces the aggregation process; instead of a global pooling layer, the enhancer combines relative information to each frame via attention leveraging the global context. Extracted dense frame-level embeddings can each represent a speaker. Thus, multiple speakers can be represented by different frame-level features in each segment. We also propose an artificially generating mixture data training framework to train the proposed HEE. Through experiments on five evaluation sets, including four public datasets, the proposed HEE demonstrates at least 10% improvement on each evaluation set, except for one dataset, which we analyse that rapid speaker changes less exist.

视听扬声器日复速度旨在检测使用听觉和视觉信号时的``谁说话。现有的视听深度数据集主要专注于会议室或新闻工作室等室内环境，这些工作室与电影，纪录片和观众情景喜剧等许多情景中的野外视频完全不同。要创建一个能够有效地比较野外视频的日复速度方法的测试平台，我们向AVA电影数据集注释说话者深度标签，并创建一个名为AVA-AVD的新基准。由于不同的场景，复杂的声学条件和完全偏离屏幕扬声器，该基准是挑战。然而，如何处理偏离屏幕和屏幕上的扬声器仍然是一个关键挑战。为了克服它，我们提出了一种新的视听关系网络（AVR-Net），它引入了有效的模态掩模，以基于可见性捕获辨别信息。实验表明，我们的方法不仅可以优于最先进的方法，而且可以更加强大，因为改变屏幕扬声器的比率。消融研究证明了拟议的AVR-NET和尤其是日复一化的模态掩模的优点。我们的数据和代码将公开可用。

扬声器日流是一个标签音频或视频录制的任务，与扬声器身份或短暂的任务标记对应于扬声器标识的类，以识别“谁谈到何时发表讲话”。在早期，对MultiSpeaker录音的语音识别开发了扬声器日益衰退算法，以使扬声器自适应处理能够实现扬声器自适应处理。这些算法还将自己的价值作为独立应用程序随着时间的推移，为诸如音频检索等下游任务提供特定于扬声器的核算。最近，随着深度学习技术的出现，这在讲话应用领域的研究和实践中引起了革命性的变化，对扬声器日益改善已经进行了快速进步。在本文中，我们不仅审查了扬声器日益改善技术的历史发展，而且还审查了神经扬声器日益改善方法的最新进步。此外，我们讨论了扬声器日复速度系统如何与语音识别应用相结合，以及最近深度学习的激增是如何引领联合建模这两个组件互相互补的方式。通过考虑这种令人兴奋的技术趋势，我们认为本文对社区提供了有价值的贡献，以通过巩固具有神经方法的最新发展，从而促进更有效的扬声器日益改善进一步进展。

Voice Conversion (VC) is the task of making a spoken utterance by one speaker sound as if uttered by a different speaker, while keeping other aspects like content unchanged. Current VC methods, focus primarily on spectral features like timbre, while ignoring the unique speaking style of people which often impacts prosody. In this study, we introduce a method for converting not only the timbre, but also prosodic information (i.e., rhythm and pitch changes) to those of the target speaker. The proposed approach is based on a pretrained, self-supervised, model for encoding speech to discrete units, which make it simple, effective, and easy to optimise. We consider the many-to-many setting with no paired data. We introduce a suite of quantitative and qualitative evaluation metrics for this setup, and empirically demonstrate the proposed approach is significantly superior to the evaluated baselines. Code and samples can be found under https://pages.cs.huji.ac.il/adiyoss-lab/dissc/ .

While recent research advances in speaker diarization mostly focus on improving the quality of diarization results, there is also an increasing interest in improving the efficiency of diarization systems. In this paper, we propose a multi-stage clustering strategy, that uses different clustering algorithms for input of different lengths. Specifically, a fallback clusterer is used to handle short-form inputs; a main clusterer is used to handle medium-length inputs; and a pre-clusterer is used to compress long-form inputs before they are processed by the main clusterer. Both the main clusterer and the pre-clusterer can be configured with an upper bound of the computational complexity to adapt to devices with different constraints. This multi-stage clustering strategy is critical for streaming on-device speaker diarization systems, where the budgets of CPU, memory and battery are tight.

In this work we propose a novel token-based training strategy that improves Transformer-Transducer (T-T) based speaker change detection (SCD) performance. The conventional T-T based SCD model loss optimizes all output tokens equally. Due to the sparsity of the speaker changes in the training data, the conventional T-T based SCD model loss leads to sub-optimal detection accuracy. To mitigate this issue, we use a customized edit-distance algorithm to estimate the token-level SCD false accept (FA) and false reject (FR) rates during training and optimize model parameters to minimize a weighted combination of the FA and FR, focusing the model on accurately predicting speaker changes. We also propose a set of evaluation metrics that align better with commercial use cases. Experiments on a group of challenging real-world datasets show that the proposed training method can significantly improve the overall performance of the SCD model with the same number of parameters.

播客本质上是对话性的，说话者的变化很频繁 - 需要说话者诊断以了解内容。我们在不依赖语言特定组件的情况下提出了一种无监督的技术诊断技术。该算法是重叠的，不需要有关说话者数量的信息。我们的方法显示，针对播客数据的Google Cloud Platform解决方案，纯度得分（F-评分为34％）的纯度得分提高了79％。

重叠的言语日期始终被视为多标签分类问题。在本文中，通过使用电源集编码多扬声器标签，我们将此任务重新格式化为单个标签预测问题。具体地，我们提出了扬声器嵌入感知的神经日复日复速节（发送）方法，其根据语音特征和给定扬声器嵌入的相似性预测电力集编码标签。我们的方法通过利用之前的文献中未能很好地研究，进一步扩展并与下游任务集成在一起。实验结果表明，我们的方法达到了比目标扬声器语音活动检测更低的日益缓释误差率。当涉及文本信息时，可以进一步降低日复速度误差。对于真正的会议场景，与基于贝叶斯隐马尔可夫模型的聚类算法相比，我们的方法可以实现相对改进34.11％。

在本文中，我们在多方会议场景中对说话者的自动语音识别（SA-ASR）进行了比较研究，这一主题越来越关注丰富的转录。具体而言，本研究评估了三种方法。第一种方法，即FD-SOT，由框架级诊断模型组成，以识别说话者和多对话者ASR以识别话语。通过对齐诊断结果和公认的假设，可以获得说话者归因的转录。但是，由于模块化的独立性，这种对齐策略可能会遭受错误的时间戳，从而严重阻碍了模型性能。因此，我们提出了第二种方法WD-SOT，以通过引入单词水平诊断模型来解决对齐误差，从而可以摆脱这种时间戳对齐依赖性。为了进一步缓解对齐问题，我们提出了第三种方法TS-ASR，该方法可以训练目标扬声器分离模块和ASR模块。通过比较每种SA-ASR方法的各种策略，对真实会议场景语料库的实验结果，AlimeTing，表明WD-SOT方法可在平均扬声器依赖性角色错误率（SD-CER）相对降低10.7％，与之相比FD-SOT方法。此外，TS-ASR方法还优于FD-SOT方法，并带来16.5％的相对平均SD-CER减少。

使用未知数量的扬声器数量的单通道远场录制的自动语音识别（ASR）传统上由级联模块解决。最近的研究表明，与模块化系统相比，端到端（E2E）多扬声器ASR模型可以实现卓越的识别准确性。但是，这些模型不会确保由于其对完整音频上下文的依赖性而实时适用性。这项工作采用实时适用性，作为模型设计的第一优先级，并解决了以前的多扬声器经常性神经网络传感器（MS-RNN-T）的几个挑战。首先，我们在训练期间介绍一般的重叠言论模拟，在LibrisPeechMix测试集上产生14％的相对字错误率（WER）改进。其次，我们提出了一种新的多转RNN-T（MT-RNN-T）模型，其具有基于重叠的目标布置策略，其概括为任意数量的扬声器，而没有模型架构的变化。我们调查在Liblics测试集上培训训练期间看到的最大扬声器数量的影响，并在两位扬声器MS-RNN-T上报告28％的相对加速。第三，我们试验丰富的转录战略，共同承认和分割多方言论。通过深入分析，我们讨论所提出的系统的潜在陷阱以及未来的未来研究方向。

The objective of this paper is speaker recognition under noisy and unconstrained conditions.We make two key contributions. First, we introduce a very large-scale audio-visual speaker recognition dataset collected from open-source media. Using a fully automated pipeline, we curate VoxCeleb2 which contains over a million utterances from over 6,000 speakers. This is several times larger than any publicly available speaker recognition dataset.Second, we develop and compare Convolutional Neural Network (CNN) models and training strategies that can effectively recognise identities from voice under various conditions. The models trained on the VoxCeleb2 dataset surpass the performance of previous works on a benchmark dataset by a significant margin.

A method to perform offline and online speaker diarization for an unlimited number of speakers is described in this paper. End-to-end neural diarization (EEND) has achieved overlap-aware speaker diarization by formulating it as a multi-label classification problem. It has also been extended for a flexible number of speakers by introducing speaker-wise attractors. However, the output number of speakers of attractor-based EEND is empirically capped; it cannot deal with cases where the number of speakers appearing during inference is higher than that during training because its speaker counting is trained in a fully supervised manner. Our method, EEND-GLA, solves this problem by introducing unsupervised clustering into attractor-based EEND. In the method, the input audio is first divided into short blocks, then attractor-based diarization is performed for each block, and finally, the results of each block are clustered on the basis of the similarity between locally-calculated attractors. While the number of output speakers is limited within each block, the total number of speakers estimated for the entire input can be higher than the limitation. To use EEND-GLA in an online manner, our method also extends the speaker-tracing buffer, which was originally proposed to enable online inference of conventional EEND. We introduce a block-wise buffer update to make the speaker-tracing buffer compatible with EEND-GLA. Finally, to improve online diarization, our method improves the buffer update method and revisits the variable chunk-size training of EEND. The experimental results demonstrate that EEND-GLA can perform speaker diarization of an unseen number of speakers in both offline and online inferences.

言语分离的许多最近进步主要针对具有高重叠程度的短音频话语的合成混合物。这些数据集与真实的会话数据显着不同，因此，在这些数据集上培训和评估的模型不会概括到真实的会话方案。使用大多数这些模型用于长形式语音的另一个问题是由于时间频率掩模或置换不变训练（PIT）损耗的无监督聚类，因此是分离的语音段的非明确顺序。这导致准确地缝合用于自动语音识别（ASR）的下游任务的均匀扬声器段。在本文中，我们提出了一种扬声器调节分离器，在直接从混合信号中提取的扬声器嵌入物上训练。我们使用定向丢失训练此模型，该丢失调节分离的段的顺序。使用此模型，我们对真实会话数据的单词错误率（WER）进行了重大改进，而无需额外的重新拼接步骤。

With the advancements in deep learning (DL) and an increasing interest in data-driven speech processing methods, there is a major challenge in accessing pathological speech data. Public challenge data offers a potential remedy for this but may expose patient health information by re-identification attacks. Therefore, we investigate in this study whether or not pathological speech is more vulnerable to such re-identification than healthy speech. Our study is the first large-scale investigation on the effects of different speech pathology on automatic speaker verification (ASV) using a real-world pathological speech corpus of more than 2,000 test subjects with various speech and voice disorders from different ages. Utilizing a DL-based ASV method, we obtained a mean equal error rate (EER) of 0.89% with a standard deviation of 0.06%, which is a factor of three lower than comparable healthy speech databases. We further perform detailed analyses of external influencing factors on ASV such as age, pathology, recording environment, utterance length, and intelligibility, to explore their respective effect. Our experiments indicate that some types of speech pathology, in particular dysphonia, regardless of speech intelligibility, are more vulnerable to a breach of privacy compared to healthy speech. We also observe that the effect of pathology lies in the range of other factors, such as age, microphone, and recording environment.

本文介绍了使用变压器基于目标扬声器语音活动检测（TS-VAD）的扬声器诊断模型。为了克服原始的TS-VAD模型无法处理任意数量的扬声器的缺点，我们研究了使用具有可变长度时间和扬声器尺寸的输入张量的模型架构。将变压器层应用于扬声器轴，以使模型输出对提供给TS-VAD模型的扬声器配置文件的顺序不敏感。时间顺序层插入了这些说话者的变压器层之间，以允许捕获输入语音信号的时间和跨语言器相关性。我们还使用基于编码器的吸引子（EEND-EDA）将基于端到端神经诊断的诊断模型通过基于变压器的TS-VAD替换其基于DOT的扬声器检测层，从而扩展了基于端到端的神经腹泻。 VoxConverse上的实验结果表明，使用变压器进行跨言扬声器建模可将TS-VAD的诊断错误率（DER）降低10.9％，从而使新的最先进（SOTA）DER达到4.74％。此外，我们的扩展eDa-eda在呼叫者数据集上相对于原始eend-eda的模型大小将6.9％降低了6.9％，在广泛使用的培训数据设置下，新的SOTA DER为11.18％。

无监督的零射声语音转换（VC）旨在修改话语的扬声器特性，以匹配看不见的目标扬声器，而无需依赖并行培训数据。最近，已经显示了语音表示的自我监督学习在不使用转录物的情况下产生有用的语言单元，这可以直接传递给VC模型。在本文中，我们展示了通过使用长度重采样解码器来实现高质量的音频样本，这使得VC模型能够与不同的语言特征提取器和声码器一起工作，而无需它们以相同的序列长度运行。我们表明，我们的方法可以胜过VCTK数据集的许多基线。在不修改架构的情况下，我们进一步展示了a）使用来自同一扬声器的不同音频段，b）添加循环一致性损失，并且c）添加扬声器分类损失可以有助于学习更好的扬声器嵌入。我们的模型使用这些技术训练了Libritts，实现了最佳性能，产生了音频样本对目标扬声器的声音，同时保留了在字符错误率方面与实际人类话语相当的语言内容。

个性化语音合成系统是一个非常期望的应用程序，其中系统可以使用罕见的登记录制与用户的语音产生语音。最近有两种主要方法可以在近期建立这样的系统：扬声器适配和扬声器编码。一方面，扬声器适配方法微调训练有素的多扬声器文本到语音（TTS）模型，只有少数注册样本。然而，它们需要至少有数千个微调步骤以进行高质量适应，使其难以在设备上施加。另一方面，扬声器编码方法将注册话语编码为扬声器嵌入。训练的TTS模型可以在相应的扬声器嵌入上综合用户的语音。然而，扬声器编码器遭受了所看到和看不见的扬声器之间的泛化差距。在本文中，我们建议将元学习算法应用于扬声器适应方法。更具体地说，我们使用模型不可知的元学习（MAML）作为多扬声器TTS模型的训练算法，其旨在找到一个很好的元初始化，以便快速地将模型调整到任何几次扬声器适应任务。因此，我们还可以将元训练的TTS模型调整为有效地解除扬声器。我们的实验比较了两个基线的提出方法（Meta-TTS）：扬声器适配方法基线和扬声器编码方法基线。评估结果表明，Meta-TTS可以从扬声器适应基线的少量适应步骤中综合高扬声器相似性语音，而不是扬声器适配基线，并且在相同的训练方案下优于扬声器编码基线。当基线的扬声器编码器用额外的8371个扬声器进行预先培训时，Meta-TTS仍然可以越优于库特布特数据集的基线，并在VCTK数据集上实现可比结果。

现有的语音克隆（VC）任务旨在将段落文本转换为具有参考音频指定的所需语音的语音。这显着提高了人工语音应用的发展。然而，也存在许多情景，这些方案不能被这些VC任务更好地反映，例如电影配音，这需要语音与与电影图一致的情绪。为了填补这个差距，在这项工作中，我们提出了一个名为Visual Voice Cloning（V2C）的新任务，该任务试图将文本段落转换为具有由参考视频指定的参考音频和所需情绪指定的所需语音的语音。为了促进该领域的研究，我们构建数据集，V2C动画，并根据现有的最先进（SOTA）VC技术提出强大的基线。我们的数据集包含10,217个动画电影剪辑，覆盖各种类型的类型（例如，喜剧，幻想）和情感（例如，快乐，悲伤）。我们进一步设计了一组名为MCD-DTW-SL的评估度量，这有助于评估地面真理语音和合成的相似性。广泛的实验结果表明，即使是SOTA VC方法也不能为我们的V2C任务产生令人满意的演讲。我们希望拟议的新任务与建设的数据集和评估度量一起将促进语音克隆领域的研究和更广泛的视野和语言社区。