语音tts已经走向了aigc大模型模式,利用大规模高品质训练数据,实现zero-shot的声音复刻正成为越来越火热的技术趋势。零样本语音合成在娱乐产业、通用ai平台及aigc领域都有广泛的应用场景,包括电影电视配音及解说、有声书、角色配音、虚拟主播、语音导航等。晴数智慧前瞻性地推出“语音复刻大模型高品质数据集”,此数据具备48khz高采样率,采集人数上万,环境纯净,内容自然多样,近万小时,是实现零样本语音合成的绝佳数据。特别地,此数据的多样化语料内容能为模型学习协同发音提供助益。
语种
中文
数据风格
对话式&朗读式
音频格式
pcm
采样率
48khz
比特率
16 bits
声道
1
人数
18,226
时长
6,692
音频数量
4,179,024个
as a new trend of aigc, zero-shot speech synthesis has wide-ranging applications, including voice assistants, audiobooks, video game character voices, creating podcast, and real-time voice changer, among others. magic data has proactively designed and developed the "high-quality dataset for voice replication," which boasts a 48khz high sampling rate, tens of thousands of contributors with diverse content, making it an excellent resource for achieving zero-shot speech synthesis. in particular, the diverse content of this data can be beneficial for the model's coarticulation learning.
符合iso/iec 27001和iso/iec 27701:2019标准认证
音频、文本、图像、音视频多模态数据
涵盖多领域的对话式、朗读式及自发式数据
人机协同高精度标注