Inside the Growing Market for AI Music Annotation
AI music development is creating demand for annotated training data, expert evaluation, and licensed music datasets, according to a 36Kr feature tracing the emerging market. MOSI.AI, the OpenMOSS team, and Shanghai Innovation Institute jointly open-sourced MOSS-Music with two 8B models and a pipeline that automates feature extraction, splitting, lyric recognition, metadata cleaning, and training-sample generation. The report also highlights MusicEval’s expert scoring dataset, Rightsify’s GCX licensing service, and Deezer’s disclosure that AI-generated tracks account for 1% to 3% of platform plays while more than 42 million such songs were excluded from recommendations between January 2025 and August 2026.
The report documents a shift from basic music tagging toward automated annotation pipelines, professional model evaluation, and rights-managed training data as AI music systems advance.
学音乐,得先听懂节奏、和弦和乐器。AI也一样。
10月,一首AI歌曲《过来人》爆火,杜甫、苏轼、刘邦等数位历史和神话人物用说唱聊房贷、职场和中年困境,戳中不少人心声。也有人直言,一听就是AI,听得浑身难受。
AI也进入了专业音乐创作舞台。江苏卫视最新的《音乐缘计划3》主打原创,卖点是原创背后的音乐故事。可第一期就有音乐人坦言作品用了AI,但能保证歌词是自己创作。
薛之谦听后却说: “词我既看得懂,又好像似曾相识的感觉。”不少观众认为,歌词是由薛之谦的几首歌融合而成。
歌本身不难听,大家的争议在于,一档主打原创价值的节目里出现了创作过程大量依赖AI的歌,凭什么和其他原创作品同台竞争?
#MOSS-Music#MOSI.AI#OpenMOSS#上海创智学院#MusicEval#ICASSP#Rightsify#GCX#Deezer#灵动音科技#UMP Board#全民K歌