Experiment resources for "Perceived Quality of Binaural Rendering From Baffled Microphone Arrays Evaluated Without an Explicit Reference"

This data set contains the following resources to reproduce the listening experiment and statistical analysis of the referenced manuscript: The binaural room impulse responses (BRIRs) of all listening conditions presented in the perceptual experiment. The tools to create the listening test infrastructure, including Pure Data (Pd) patches and configuration files for the SoundScape Renderer (SSR) and graphical user interface (GUI). The raw response data as gathered from the experiment subjects. The R and Stan scripts to perform the statistical analysis and generate the resulting plots and tables. The archive contains the following components described below. Directory dependencies/: Matlab, R, Stan, and Pd functions that are utilized in the code and experimental setup Additional dependencies of available open-source projects may be required for certain code functions. If so, the source and setup process for the necessary dependencies are documented in the file header. Directory plots/4_Equalization/: Plots of all available headphone equalization filters as generated by the following Matlab scripts. Directory plots/8_User_study/: Plots of the raw and analyzed experimental results as generated by the following Matlab and R scripts. Directory resources/BRIR_auralization/: Audio files with static binaural auralizations of all listening conditions presented in the perceptual experiment as generated by the following Matlab scripts. The files in Kemar_HRTF_sofa_N44_adjusted/ do not include a headphone equalization. The files in Kemar_HRTF_sofa_N44_adjusted+Sennheiser_HD650_lin/ include the equalization for the Sennheiser HD650 headphones employed in the listening experiment. The files are identical to the annotated listening examples published for the manuscript. Directory resources/BRIR_rendered/: BRIRs and rendering parameters of all listening conditions presented in the perceptual experiment as generated from the associated data set and rendering code. Scene configuration files for the SSR with all listening conditions presented in the perceptual experiment as generated from the following Matlab scripts. Directory resources/HPCF_KEMAR/: Impulse responses of equalization filters for various headphones on the G.R.A.S KEMAR acoustic dummy head as measured for this experiment. Directory resources/User_study/: Various resources for the listening experiment. The files in Exp1_analysis/ include intermediate and final statistical analysis results as generated from the following R scripts. Exp1_config.json contains the configuration of the study GUI with conditions presented in the listening experiment. Exp1_Introduction.pdf contains the instructions presented to the subjects at the start of the listening experiment. Exp1_Part2_data_strings.xls contains all subjects' raw perceptual response data gathered from the listening experiment. Questionnaire.pdf contains the questionnaire given to the subjects at the end of the listening experiment. Matlab script x4_Gather_Headphone_Compensations.m: Generate plots of measured headphone compensation filters. Furthermore, the generated minimum phase filters and filters to yield a linear phase response from the headphones are extracted as separate WAV files. Matlab script x6_Gather_SSR_Configurations.m: Collect several specified pre-rendered binaural room impulse response sets into an ASD file. The SSR can load this scene to present all gathered configurations in direct comparison with head tracking. Readme file x6a_Normalize_SSR_Loudnesses.txt: Ideally, the rendering script would implement a measure to provide a reliable estimation of the binaural loudness of the rendered configuration. This could be used to normalize all stimuli levels. However, such a measure is currently not available or implemented. Therefore, tuning the stimuli loudness for the user study was performed beforehand by ear. The adjusted playback levels are set in a modified SSR configuration file for the listening experiment. Matlab script x6_Gather_SSR_Configurations.m: Perform convolution of (rendered) binaural room impulse responses with a source audio signal. This is done for a specified selection of static head orientations and a continuous rotation over all horizontal head orientations. The resulting auralizations are published as supplementary materials to the manuscript. Shell script x7_Start_Study_GUI.sh: Initialize all required components to perform the perceptual user study, including: SSR to perform the real-time rendering of the BRIRs with head tracking SSR to extract head-tracking data (in case a Polhemus tracker is used) Pd to extract head-tracking data (in case a Supperware tracker is used) Pd to perform real-time convolution to apply headphone compensation Pd to receive OSC messages from the study GUI Pd to trigger audio file playback from received OSC messages Pd to translate OSC messages into FUDI messages for the SSR The GUI to be used by the participants and implement the study procedure Some static configuration variables can be adjusted, whereas other parameters are chosen during script execution. Matlab script x8_Gather_Study_Data.m: Transform the raw result data from the questionnaire (*.xls) and the study GUI (*.json) into a compact format (*.xls) that can be loaded to plot the raw data and imported by software for the subsequent statistical analysis. Note that this contains the responses from all subjects, whereas responses from the investigators must be excluded from the statistical analysis (which is implemented in the analysis scripts). Matlab script x8a_Plot_Study_Data.m: Generate a set of violin plots to visualize the initial distribution of the raw perceptual data. The data is split by specified attributes and plotted separately for visual inspection. The data may also be transformed into ranks for a first distribution inspection. Note that implementing the ranking method, notably how ties are resolved, may differ from the technique employed in the statistical analysis. Note that this contains the responses from all subjects, whereas responses from the investigators must be excluded from the statistical analysis (which is implemented in the analysis scripts). R script x8b_Analyze_Exp1_Data.R: Perform the statistical analysis by transforming the observed subject ratings into a predicted distribution of ranks using a hierarchical generalized linear regression model. Executing the statistical model may take some time due to the Bayesian framework employing Markov-chain Monte Carlo simulations. Data is exported at various intermediate steps to be loaded and visualized by the following R script. R markdown script x8c_Plot_Exp1_Results.Rmd: Generate various plots and data tables of the observed data and the predicted results to visualize the distribution and influence of different analysis parameters. Some of the resulting plots were used in the manuscript. x8c_Plot_Exp1_Results.html conveniently summarizes all plots and data tables generated by "knitting" the R markdown script.本数据集包含以下资源，用于复现参考文献稿件中的听觉实验与统计分析工作： 1. 感知实验中所有听觉条件下的双耳房间冲激响应（binaural room impulse responses, BRIRs）。 2. 用于搭建听觉测试平台的工具，包括纯数据（Pure Data, Pd）补丁文件，以及声景渲染器（SoundScape Renderer, SSR）的配置文件与图形用户界面（graphical user interface, GUI）。 3. 从实验受试者处收集到的原始响应数据。 4. 用于执行统计分析、生成结果图表与表格的R与Stan脚本。 本存档包含如下所述的多个组件。 目录`dependencies/`： 包含代码与实验配置中用到的MATLAB、R、Stan及Pd函数。 部分代码功能可能需要额外的开源项目依赖，相关依赖的获取与安装流程已在对应文件的头部进行说明。 目录`plots/4_Equalization/`： 包含由下述MATLAB脚本生成的所有可用耳机均衡滤波器的绘图结果。 目录`plots/8_User_study/`： 包含由下述MATLAB与R脚本生成的实验原始结果与分析结果绘图。 目录`resources/BRIR_auralization/`： 包含由下述MATLAB脚本生成的、对应感知实验中所有听觉条件的静态双耳听觉化音频文件。 其中`Kemar_HRTF_sofa_N44_adjusted/`目录下的文件未应用耳机均衡处理。 `Kemar_HRTF_sofa_N44_adjusted+Sennheiser_HD650_lin/`目录下的文件则应用了本听觉实验中使用的森海塞尔HD650耳机的均衡处理，与该稿件配套发布的标注听觉示例完全一致。 目录`resources/BRIR_rendered/`： 包含由关联数据集与渲染代码生成的、对应感知实验中所有听觉条件的双耳房间冲激响应与渲染参数。 同时包含由下述MATLAB脚本生成的、适用于声景渲染器的所有听觉条件场景配置文件。 目录`resources/HPCF_KEMAR/`： 包含本实验中测量得到的、搭载于G.R.A.S KEMAR声学仿真人头模型上的多款耳机的均衡滤波器冲激响应。 目录`resources/User_study/`： 包含听觉实验的各类配套资源。 `Exp1_analysis/`目录下的文件包含由下述R脚本生成的中间统计分析结果与最终统计分析结果。 `Exp1_config.json`包含本听觉实验中采用的实验GUI配置与听觉条件信息。 `Exp1_Introduction.pdf`包含听觉实验开始前向受试者展示的实验说明。 `Exp1_Part2_data_strings.xls`包含从听觉实验中收集到的所有受试者的原始感知响应数据。 `Questionnaire.pdf`包含听觉实验结束后向受试者发放的调查问卷。 MATLAB脚本`x4_Gather_Headphone_Compensations.m`： 用于生成实测耳机补偿滤波器的绘图结果，同时将生成的最小相位滤波器与用于将耳机响应转换为线性相位的滤波器提取为独立的WAV格式文件。 MATLAB脚本`x6_Gather_SSR_Configurations.m`： 1. 将多个指定的预渲染双耳房间冲激响应集合整合为ASD文件，声景渲染器可加载该场景文件，实现所有采集配置的实时对比并支持头部追踪。 2. 对（预渲染的）双耳房间冲激响应与源音频信号执行卷积操作，该操作针对指定的静态头部朝向集合，以及覆盖所有水平头部朝向的连续旋转场景。 生成的听觉化结果已作为该稿件的补充材料发布。 README文件`x6a_Normalize_SSR_Loudnesses.txt`： 理想情况下，渲染脚本应实现一种可可靠估算渲染配置双耳响度的方法，用于归一化所有刺激声的声级，但目前尚无此类可用的实现方案。 因此，本用户研究中的刺激声声级调节已预先通过试听完成，调整后的播放电平已在针对听觉实验修改的SSR配置文件中设置。 Shell脚本`x7_Start_Study_GUI.sh`： 用于初始化执行感知用户研究所需的所有组件，包括： - 支持头部追踪的BRIR实时渲染的声景渲染器 - 用于提取头部追踪数据的声景渲染器（若使用Polhemus追踪器） - 用于提取头部追踪数据的Pd程序（若使用Supperware追踪器） - 用于执行实时卷积以应用耳机补偿的Pd程序 - 用于接收来自实验GUI的开放声音控制（Open Sound Control, OSC）消息的Pd程序 - 用于根据接收的OSC消息触发音频文件播放的Pd程序 - 用于将OSC消息转换为适用于声景渲染器的流体用户界面（Fluid User Interface, FUDI）消息的Pd程序 - 供受试者使用的实验GUI，用于实现实验流程 部分静态配置变量可自行调整，其余参数则在脚本执行过程中选定。 MATLAB脚本`x8_Gather_Study_Data.m`： 用于将调查问卷（*.xls）与实验GUI（*.json）导出的原始结果数据转换为紧凑格式（*.xls），该格式可用于加载绘制原始数据，也可导入软件以执行后续统计分析。 请注意，该文件包含所有受试者的响应数据，统计分析脚本已实现排除研究者自身响应数据的逻辑。 MATLAB脚本`x8a_Plot_Study_Data.m`： 用于生成一组小提琴图（violin plots），以可视化原始感知数据的初始分布。数据将按指定属性进行拆分并分别绘图，便于目视检查。 数据也可转换为秩数据以进行初步分布检验。请注意，秩转换方法（尤其是结值处理方式）可能与统计分析中采用的技术存在差异。 请注意，该文件包含所有受试者的响应数据，统计分析脚本已实现排除研究者自身响应数据的逻辑。 R脚本`x8b_Analyze_Exp1_Data.R`： 通过将受试者的观测评分转换为秩的预测分布，利用层次广义线性回归模型（hierarchical generalized linear regression model）执行统计分析。 由于该统计模型采用马尔可夫链蒙特卡洛模拟（Markov-chain Monte Carlo simulations）的贝叶斯框架，执行过程可能耗时较长。 数据将在多个中间步骤导出，以供下述R脚本加载并可视化。 R Markdown（R Markdown）脚本`x8c_Plot_Exp1_Results.Rmd`： 用于生成观测数据与预测结果的各类绘图与数据表格，以可视化不同分析参数的分布与影响。 其中部分生成的绘图已用于该稿件中。 `x8c_Plot_Exp1_Results.html`通过“编织”该R Markdown脚本生成，可便捷地汇总所有绘图与数据表格结果。