管廊管线故障声纹(MFCC)数据集
收藏资源简介：
管廊管线故障声纹库数据在试点管廊（容西管廊豪丹段综合仓与讯电仓；容东管廊205段能源仓，水信仓与高压仓；容东管廊207段水信仓）采集与实验室管廊试验台中采集。在试点管廊中使用声纹传感器与录音主机采集管廊中的巡检脚步声纹与风机工作时的声纹；在北京交通大学唐山研究院载运工具运行安全实验室中通过实验平台采集管廊失效相关的声纹（管廊供电线路失效声纹；管廊供水管路泄漏声纹；管廊输气线路泄漏声纹；管廊供电线路失效与供水管路泄漏组合声纹；管廊供电线路失效与输气线路泄漏组合声纹；管廊供水管路泄漏与输气线路泄漏组合声纹）。通过模拟实际管廊的结构和尺寸比例，设计管廊数据采集实验平台，在管廊试验台中布置管线和采集设备，模拟不同工况下管线失效产生的声纹。 所有采集的声学数据均保存为WAV格式，以保证音频数据的高保真度，避免因压缩而造成的频谱信息丢失。采集过程中严格记录每一条数据的元信息，包括声源类型、泄漏位置、泄漏尺寸、管道压力、采集距离等。试验完成后，所有音频数据将被人工标注并存储到管廊声纹数据库中。在标注阶段，还将对音频进行时间切片处理，将每段音频分割为1秒时长的文件。同时对采集到的数据需要进行人工筛选，去掉不合格的音频和音频中空白的部分，保证音频的质量。 在完成WAV格式的音频数据集处理和建立后，需要将音频数据集进行预处理提取特征转换成MFCC梅尔谱图，然后对应保存成MFCC数据集。这个过程主要包括分帧，加窗，短时傅里叶变换和梅尔倒谱分析。 管廊管线故障音频MFCC数据集包含9个类别，分别是供水管道泄漏声纹，供气管道泄漏声纹，供电线路泄漏声纹、巡检脚步声纹、排风机声纹、供水管道泄漏与供气管道泄漏组合声纹、供水管道泄漏与供电线路泄漏组合声纹、供气管道泄漏与供电线路泄漏组合声纹、供水管道泄漏与供气管道泄漏与供电线路泄漏组合声纹，通过对WAV格式数据进行预处理提取特征转换后，然后按照每一帧进行切割得到，共940800条数据，1.66GB。
The acoustic fingerprint database data for utility tunnel pipeline faults was collected from both pilot utility tunnels and laboratory utility tunnel test benches. The pilot utility tunnels include: Haodan Section Integrated Warehouse and Telecommunication-Electricity Warehouse of Rongxi Utility Tunnel; Energy Warehouse, Water-Telecommunication Warehouse and High-Pressure Warehouse of Rongdong Utility Tunnel Section 205; Water-Telecommunication Warehouse of Rongdong Utility Tunnel Section 207. In the pilot utility tunnels, acoustic fingerprint sensors and recording hosts were used to collect the acoustic fingerprints of inspection footsteps and fan operation sounds in the tunnels. In the Vehicle Operation Safety Laboratory of Tangshan Research Institute, Beijing Jiaotong University, acoustic fingerprints related to utility tunnel failures were collected via an experimental platform, including: acoustic fingerprints of power supply line faults, acoustic fingerprints of water supply pipeline leaks, acoustic fingerprints of gas transmission pipeline leaks, combined acoustic fingerprints of power supply line faults and water supply pipeline leaks, combined acoustic fingerprints of power supply line faults and gas transmission pipeline leaks, and combined acoustic fingerprints of water supply pipeline leaks and gas transmission pipeline leaks. A utility tunnel data acquisition experimental platform was designed by simulating the actual structure and dimensional proportions of utility tunnels, with pipelines and acquisition equipment arranged in the utility tunnel test bench to simulate the acoustic fingerprints generated by pipeline failures under different working conditions. All collected acoustic data is saved in WAV format to ensure high fidelity of the audio data and avoid loss of spectral information caused by compression. During the acquisition process, the metadata of each data record is strictly recorded, including sound source type, leak location, leak size, pipeline pressure, acquisition distance, etc. After the completion of the experiment, all audio data will be manually annotated and stored in the utility tunnel acoustic fingerprint database. During the annotation stage, time slicing processing will also be performed on the audio, dividing each segment of audio into 1-second long files. Meanwhile, manual screening of the collected data is required to eliminate unqualified audio and silent parts in the audio, ensuring the quality of the audio. After the completion of the processing and establishment of the WAV-format audio dataset, preprocessing is required to extract features and convert them into MFCC (Mel-Frequency Cepstral Coefficients) mel-spectrograms, which are then saved as the MFCC dataset. This process mainly includes framing, windowing, Short-Time Fourier Transform (STFT) and mel-frequency cepstral analysis. The MFCC dataset for utility tunnel pipeline fault audio contains 9 categories, which are: water supply pipeline leak acoustic fingerprint, gas supply pipeline leak acoustic fingerprint, power supply line leak acoustic fingerprint, inspection footstep acoustic fingerprint, exhaust fan acoustic fingerprint, combined acoustic fingerprint of water supply pipeline leak and gas supply pipeline leak, combined acoustic fingerprint of water supply pipeline leak and power supply line leak, combined acoustic fingerprint of gas supply pipeline leak and power supply line leak, and combined acoustic fingerprint of water supply pipeline leak, gas supply pipeline leak and power supply line leak. The dataset is obtained by preprocessing the WAV-format data to extract features and convert them, then cutting according to each frame, with a total of 940,800 data entries and a size of 1.66 GB.
