多场景对讲机语音指令识别与降噪优化数据集

本数据集为企业自主生产，通过 “设备采集 + 人工标注 + 算法处理” 全流程构建：语音指令基础数据：组织专业人员在无噪环境下录制普通话多语速（快速、中等、慢速）及安徽多地方言的对讲机核心指令（如 “呼叫群组”“紧急撤离” 等），人工标注口音、语速、清晰度等参数； 特殊场景噪音数据：在城市街道（车流噪音）、工地（机械噪音）、人群密集区（商场 / 车站人声）、风雨天气等实地环境中，采集带噪音的语音指令，通过自研降噪算法处理后，标注原始 / 降噪后清晰度、识别准确率； 全流程执行《数据采集与标注 SOP》，确保数据结构统一参数准确，构成 “语音指令 - 场景- 处理结果” 的完整数据集合。This dataset is independently developed by the enterprise and constructed through the end-to-end workflow of "device collection + manual annotation + algorithm processing": Basic voice command data: Professional staff were organized to record core walkie-talkie commands in Mandarin with three speaking speeds (fast, medium, slow) and various Anhui local dialects in a noise-free environment, such as "call group", "emergency evacuation", etc. Parameters including accent, speaking speed and intelligibility were manually annotated; Special scenario noise-contaminated voice command data: Voice commands with background noise were collected in real-world environments including urban streets (traffic noise), construction sites (mechanical noise), crowded areas (mall/station crowd voices), and windy/rainy weather. After being processed by the self-developed noise reduction algorithm, the original and denoised intelligibility and recognition accuracy were annotated; The entire workflow strictly complies with the Data Collection and Annotation SOP to ensure unified data structure and accurate parameters, forming a complete dataset of "voice command - scenario - processing result".

搜集汇总

背景与挑战

背景概述

该数据集专注于对讲机语音指令识别与降噪优化，包含250条每日更新的语音数据，涵盖普通话标准语速无噪指令如‘呼叫群组 1’和‘紧急呼叫’，并标注了口音、语速、噪音和准确率等参数。它设计用于多场景环境（如城市街道和工地），支持安徽方言变体，通过结合小波变换降噪和深度神经网络识别模型，旨在提升对讲机在复杂噪音下的语音识别性能。