Self-supervised learning (SSL) methods such as WavLM have shown promising speech separation (SS) results in small-scale simulation-based experiments. In this work, we extend the exploration of the SSL-based SS by massively scaling up both the pre-training data (more than 300K hours) and fine-tuning data (10K hours). We also investigate various techniques to efficiently integrate the pre-trained model with the SS network under a limited computation budget, including a low frame rate SSL model training setup and a fine-tuning scheme using only the part of the pre-trained model. Compared with a supervised baseline and the WavLM-based SS model using feature embeddings obtained with the previously released 94K hours trained WavLM, our proposed model obtains 15.9% and 11.2% of relative word error rate (WER) reductions, respectively, for a simulated far-field speech mixture test set. For conversation transcription on real meeting recordings using continuous speech separation, the proposed model achieves 6.8% and 10.6% of relative WER reductions over the purely supervised baseline on AMI and ICSI evaluation sets, respectively, while reducing the computational cost by 38%.
translated by 谷歌翻译
We present the first neural network model to achieve real-time and streaming target sound extraction. To accomplish this, we propose Waveformer, an encoder-decoder architecture with a stack of dilated causal convolution layers as the encoder, and a transformer decoder layer as the decoder. This hybrid architecture uses dilated causal convolutions for processing large receptive fields in a computationally efficient manner, while also benefiting from the performance transformer-based architectures provide. Our evaluations show as much as 2.2-3.3 dB improvement in SI-SNRi compared to the prior models for this task while having a 1.2-4x smaller model size and a 1.5-2x lower runtime. Open-source code and datasets: https://github.com/vb000/Waveformer
translated by 谷歌翻译
本文介绍了一个新型的流媒体自动语音识别（ASR）框架，用于由带有任意几何形状的遥远麦克风阵列捕获的多对话者重叠语音。我们的名为T-Sot-VA的框架在独立开发了两种最近的技术上。基于令牌级别的序列化输出训练（T-SOT），数量几何形状 - 反应连续的语音分离或VARARRARY和流媒体多对话者ASR。为了结合两种技术的最佳，我们新设计了一个基于T-SOT的ASR模型，该模型基于Vararray的两个分离的语音信号生成序列化的多对话者转录。我们还为这种ASR模型提出了一种预训练方案，我们基于单膜单键式ASR训练数据来模拟Vararray的输出信号。使用AMI会议语料库的对话转录实验表明，基于提议的框架的系统大大优于常规的框架。我们的系统分别在保留流媒体推理能力的同时，在多远离微米频道设置中分别实现了AMI开发和评估集的最新单词错误率为13.7％和15.5％。
translated by 谷歌翻译
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％。
translated by 谷歌翻译
本文介绍了流式扬声器的自动语音识别（SA-ASR）模型，该模型可以识别``即使多个人同时讲话，谁说'谁说什么”。我们的模型基于令牌级的序列化输出培训（T-SOT），该培训最近提议以流媒体方式转录多对词的演讲。为了进一步认识说话者的身份，我们提出了一个基于编码器的扬声器嵌入提取器，该扬声器可以估算每个公认的代币的说话者表示，不仅是从非重叠的语音中，而且还来自重叠的语音。所提出的扬声器嵌入为T-vector，与T-SOT ASR模型同步提取，从而可以通过低潜伏期的多词器转录来联合执行说话者识别（SID）或说话者诊断（SD）。我们通过使用LibrisPeechMix和Libralics Corpora评估了ASR和SID/SD联合任务的建议模型。所提出的模型比以前的流媒体模型获得了更高的准确性，并且与最新的离线SA-ASR模型显示出可比甚至更高的结果。
translated by 谷歌翻译
本文提出了代币级别的序列化输出训练（T-SOT），这是流式传输多对话者自动语音识别（ASR）的新型框架。与使用多个输出分支的现有流媒体多对话者ASR模型不同，T-SOT模型只有一个单个输出分支，该分支基于其排放时间生成多个扬声器的识别令牌（例如，单词，子字）。引入了指示“虚拟”输出通道更改的特殊令牌，以跟踪重叠的话语。与先前的流媒体ASR模型相比，T-SOT模型具有较低的推理成本和更简单的模型体系结构的优点。此外，在我们对LibrisPeechMix和Librics数据集的实验中，基于T-SOT的变压器换能器模型可实现最新的单词错误率，从而有很大的差距。对于非重叠的语音，T-SOT模型在精度和计算成本方面与单调的ASR模型相提并论，为单个单词和多对话者方案部署一个模型打开了大门。
translated by 谷歌翻译
自我监督学习（SSL）在语音识别方面取得了巨大的成功，而有限的探索已尝试完成其他语音处理任务。由于语音信号包含多方面的信息，包括说话者身份，副语言学，口语内容等，学习所有语音任务的通用表示都具有挑战性。为了解决该问题，我们提出了一个新的预培训模型WAVLM，以解决全堆栈的下游语音任务。 Wavlm共同学习了蒙面的语音预测和预训练。通过这种方式，WAVLM不仅可以通过掩盖的语音预测来保持语音内容建模能力，而且还可以通过语音denoing来提高非ASR任务的潜力。此外，WAVLM还采用封闭式的变压器结构的封闭相对位置偏置，以更好地捕获输入语音的序列排序。我们还将培训数据集从60k小时扩展到94K小时。 WAVLM大型在精湛的基准上实现了最先进的性能，并在其代表性基准上为各种语音处理任务带来了重大改进。代码和预培训模型可在https://aka.ms/wavlm上找到。
translated by 谷歌翻译
Intelligent agents have great potential as facilitators of group conversation among older adults. However, little is known about how to design agents for this purpose and user group, especially in terms of agent embodiment. To this end, we conducted a mixed methods study of older adults' reactions to voice and body in a group conversation facilitation agent. Two agent forms with the same underlying artificial intelligence (AI) and voice system were compared: a humanoid robot and a voice assistant. One preliminary study (total n=24) and one experimental study comparing voice and body morphologies (n=36) were conducted with older adults and an experienced human facilitator. Findings revealed that the artificiality of the agent, regardless of its form, was beneficial for the socially uncomfortable task of conversation facilitation. Even so, talkative personality types had a poorer experience with the "bodied" robot version. Design implications and supplementary reactions, especially to agent voice, are also discussed.
translated by 谷歌翻译
Generative models, particularly GANs, have been utilized for image editing. Although GAN-based methods perform well on generating reasonable contents aligned with the user's intentions, they struggle to strictly preserve the contents outside the editing region. To address this issue, we use diffusion models instead of GANs and propose a novel image-editing method, based on pixel-wise guidance. Specifically, we first train pixel-classifiers with few annotated data and then estimate the semantic segmentation map of a target image. Users then manipulate the map to instruct how the image is to be edited. The diffusion model generates an edited image via guidance by pixel-wise classifiers, such that the resultant image aligns with the manipulated map. As the guidance is conducted pixel-wise, the proposed method can create reasonable contents in the editing region while preserving the contents outside this region. The experimental results validate the advantages of the proposed method both quantitatively and qualitatively.
translated by 谷歌翻译
给定数千种同样准确的机器学习（ML）模型，用户如何在其中选择？最近的ML技术使领域专家和数据科学家能够为稀疏决策树生成完整的Rashomon设置，这是一套几乎最理想的可解释的ML模型。为了帮助ML从业者识别具有此Rashomon集合中理想属性的模型，我们开发了Timbertrek，这是第一个交互式可视化系统，该系统总结了数千个稀疏决策树的规模。两种用法方案突出了Timbertrek如何使用户能够轻松探索，比较和策划与域知识和价值观保持一致的模型。我们的开源工具直接在用户的计算笔记本和Web浏览器中运行，从而降低了创建更负责任的ML模型的障碍。Timbertrek可在以下公共演示链接中获得：https：//poloclub.github.io/timbertrek。
translated by 谷歌翻译