Gamingdeputy reported on April 17 that Kunlun Wanwei announced today that the performance of the Tiangong 3.0 large model has been significantly improved, and its Tiangong SkyMusic large music model is also open to the whole society for public testing today.
Tiangong 3.0 has 400 billion parameters, surpassing Grok-1 with 314 billion parameters.The world's largest open source MoE model. Tiangong 3.0 has significantly improved performance in areas such as semantic understanding, logical reasoning, versatility, generalization, uncertainty knowledge, and learning capabilities, and its mathematics/reasoning/coding/cultural and creative abilities have improved by more than 30%. Tiangong 3.0 adds multiple AI capabilities such as multi-round search and comprehensive tool invocation, chart drawing, research mode, enhancement mode, and map modification and expansion.
Advertisement
Tiangong SkyMusic music model, a subsidiary of Tiangong 3.0, is also open to the public for public testing today. Kunlun Wanwei said that Tiangong SkyMusic is “significantly” ahead of its competitors in areas such as vocal & BGM sound quality, vocal naturalness, and pronunciation intelligibility.Overall performance surpasses Suno V3obtain the large music model SOTA (State of the art model, the model that performs best in the current research).
SkyMusic adopts a Sora-like model architecture in the field of music and audio. The Large-scale Transformer is responsible for composing music to learn the contextual dependencies of Music Patches while achieving music controllability. The Diffusion Transformer is responsible for singing and using LDM to restore Music Patches to high-quality music. quality audio, enabling it to support generating 80 seconds 44100Hz sampling rate two-channel stereo song.
Advertisement
According to reports, Tiangong SkyMusic has the following characteristics:
High-quality AI music: Generate 80 seconds 44100Hz sampling rate two-channel stereo AI song
The human voice is “fake and real”: the Chinese level is extremely good, the pronunciation is clear and there is no abnormal sound
Lyric section control: The generated songs can clearly distinguish the emotional changes of different lyric sections.
Multiple music styles: support rap/folk/funk/antique/electronic, etc.
Intelligent expression of music: Able to learn various singing techniques such as vibrato, opera, singing, male and female duet, automatic harmony, etc.
Reference music generation: Users upload their own reference music to generate songs with similar styles and vocals.
Dialect song generation: supports Cantonese, Chengdu dialect, Beijing dialect and many other dialects
Gamingdeputy learned from public information that Kunlun Wanwei is a Chinese Internet platform company that has been deeply involved in overseas markets for more than ten years. Its business covers information distribution, social networking, entertainment, metaverse, games and AIGC and other fields. Its subsidiaries include AGI and The three major business sectors include AIGC, overseas information distribution and Yuanverse, and investment, with markets covering China, Southeast Asia, Africa, the Middle East, North America, South America, Europe and other places. As of now, the global average monthly active users are nearly 400 million, and overseas revenue accounts for 84%.