Multidimensional analysis of simulated conversations

在之前的研究项目中(项目编号：320253669)，在存在丢包和传输延迟的情况下，开发了一个会话模拟，模拟了轮流和语音可理解性。由此产生的对话被用来预测整体的对话质量。然而，对话质量可以分为听、说和对话交互阶段的感知维度，以提供对质量评估的更深入理解，并提供导致退化的潜在线索。ITU-T建议P.804对诊断测试程序进行了标准化，以捕捉三个对话阶段及其感知维度。本研究项目的目的是根据ITU-T建议P.804模拟听、说和交互阶段，并利用所得数据预测感知维度。为此，在仿真环境中实现了三个会话阶段的测试协议。对于语音阶段的模拟，开发了用于检测自身语音反馈（回声）的模型，以及用于在存在这种退化的情况下适应说话行为的模型。对于听力阶段的尺寸预测，可以使用正在标准化过程中的P.AMD模型。为了预测说话和互动阶段的感知维度，正在开发和评估新的模型。由于这个研究项目的结果，一个新的模拟和质量预测模型受损的对话有望。该模型和仿真应作为国际标准推广，并传递给行业合作伙伴。

In the preceding research project (project no. 320253669), a simulation of conversations was developed that modeled turn taking and speech intelligibility in the presence of packet loss and transmission delays. The resulting conversations were used to predict overall conversation quality. However, conversation quality can be divided into perceptual dimensions of the listening, speaking, and interaction phases of a conversation, to provide a deeper understanding of the quality assessment and potential clues to the cause of the degradations. ITU-T Recommendation P.804 standardizes the diagnostic testing procedure to capture the three conversational phases with their perceptual dimensions. It is the aim of this research project to simulate the listening, speaking and interaction phases according to ITU-T Recommendation P.804 and to predict the perceptual dimensions using the resulting data. For this purpose, the test protocol of the three conversation phases is implemented in the simulation environment. For the simulation of the speech phase, models are developed for the detection of the feedback of own speech (echo) and also for the adaptation of the speaking behavior in the presence of this degradation. For the prediction of the dimensions of the listening phase, the model P.AMD, which is in the process of standardization, can be used. For the prediction of the perceptual dimensions of the speaking and interaction phases, new models are being developed and evaluated. As a result of this research project, a new simulation as well as a quality prediction model for impaired conversations are expected. The model and the simulation should be promoted as an international standard, and be transferred to the industry partner.