一种听歌识人方法、终端设备及存储介质
By using self-supervised learning models and acoustic feature extraction technology, the problems of high data collection costs and low recognition accuracy in singer identification by listening to songs have been solved, achieving low-cost and high-accuracy singer identification.
CN115862642BActive Publication Date: 2026-07-17XIAMEN KUAISHANGTONG TECH CORP LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIAMEN KUAISHANGTONG TECH CORP LTD
- Filing Date
- 2021-09-24
- Publication Date
- 2026-07-17
Smart Images
Figure CN115862642B_ABST
Abstract
本发明涉及一种听歌识人方法、终端设备及存储介质，该方法中包括：采集单一说话人的音频构建第一训练集；构建基于卷积神经网络和残差神经网络的自监督学习模型；基于第一训练集中的音频的声学特征，对模型进行n次回归训练；基于第一训练集中的各音频，对模型进行第n+1次回归训练；基于第n+1次回归训练结果构建鉴别任务，将鉴别训练后的模型作为听歌识人模型；采集标注有歌手名音频数据对听歌识人模型进行分类训练，得到最终听歌识人模型；通过最终听歌识人模型对待识别歌曲的歌手进行识别。本发明无需支付高额版权费收集大量歌手歌曲，也无需人工对歌曲进行截取，只需要适量正常的带说话人标注的语音及歌手歌曲片段即可实现对听歌识人模型的训练。
Need to check novelty before this filing date? Find Prior Art