移动端小云小云唤醒词正样本测试集

针对语音唤醒任务中的“小云小云”模型，使用移动端设备(安卓手机)录制正样本测试集。测试集录制共分为9个场景，每个场景录制50句，每一句的标签均为“小云小云”。说话人音量统一校准成57dB，同步播放不同类型和信噪比的噪声，从音频文件命名可以辨别。 负样本测试集暂未开放，使用者可以根据项目使用场景自己录制数十个小时的噪声数据用于模型误唤醒性能的测试。

For the "Xiaoyun Xiaoyun" wake-word model in the speech wake-up task, the positive sample test set was recorded using mobile devices (Android smartphones). The test set recording includes 9 scenarios, with 50 utterances recorded in each scenario, and each utterance is uniformly labeled as "Xiaoyun Xiaoyun". The speaker's volume was uniformly calibrated to 57 dB, and noises of various types and signal-to-noise ratios (SNR) were played synchronously, which can be identified from the audio file names. The negative sample test set is not yet open to users. Users can record dozens of hours of noise data according to the actual usage scenarios of their projects to test the model's false wake-up performance.

搜集汇总

背景与挑战

背景概述

该数据集是专为移动端'小云小云'语音唤醒模型设计的正样本测试集，包含在9种不同噪声场景下录制的音频，每种场景有50条标注为'小云小云'的语音，说话人音量统一为57dB并伴有多种噪声。它用于测试唤醒模型的性能，支持唤醒率评估，音频格式为16KHz单声道。