Since the beginning of this year, with the launch of a number of "AI song-writing artifacts" such as Suno and Udio, the research and development of AI large models and the discussion of application scenarios have gradually shifted to the music field.
In the domestic market, Kunlun Wanwei's first domestic music SOTA model "Tiangong SkyMusic" has also become the focus of heated discussions in the industry.
As the first publicly available AI music model in China, "Tiangong SkyMusic" has ignited the creative enthusiasm of the whole people and stimulated the market's high attention to the concept of "AI music", making Kunlun Wanwei, as an A-share listed company, also repeatedly searched for "AI music concept stocks". Weibo address: https://weibo.com/u/7891361412
The strength of "SkyMusic" should not be underestimated. In the horizontal evaluation with Suno V3, the top overseas AI music model, "Tiangong SkyMusic" is significantly ahead of its competitors in the fields of vocal & BGM sound quality, vocal naturalness, pronunciation intelligibility, etc., and has become a global AI music SOTA model.
In the highly competitive AI music model industry, how can "SkyMusic" stand out?
On April 18th, Fang Han, chairman and CEO of Kunlun Wanwei, was invited by Qubit to share a live broadcast with the title of "China's First Music SOTA Model Shocking Debut: Kunlun Wanwei Tiangong 3.0 Ultimate Revelation", which is the first time that Fang Han has comprehensively interpreted China's first music SOTA model "Tiangong SkyMusic" and "Tiangong 3.0" for the majority of users.
Perhaps Fang Han's live broadcast sharing may be able to reveal the answers to the above questions for us.
Fang Han mentioned that the success of the "Tiangong SkyMusic" music model is inseparable from Kunlun's technical accumulation in the field of music over the years. Over the past three years, the SkyMusic team has overcome many AI music technology challenges.
Originally, the research route of "Tiangong SkyMusic" focused on the symbolic music generation technology, and although the quality and effect of the music have reached the excellent level, it is difficult to scale up the technical route.
Therefore, the team decided to completely reinvent the previous work and adopt the self-developed large-scale model music audio generation technology route, which directly realizes the integrated end-to-end music generation of instruments, vocals, melodies, volumes, and notes through large-scale model technology.
Despite the technical difficulties, the SkyMusic team did not back down. Through in-depth research on large-scale model technology, continuous optimization of algorithms, and overcoming one technical difficulty after another, we finally succeeded in achieving a breakthrough in large-scale model music audio generation technology and reached the leading level in the industry.
In addition to the in-depth interpretation of "Tiangong SkyMusic", Fang Han also shared the wonderful songs created by users through "Tiangong SkyMusic" in real time at the live broadcast site, among them, there are not only family moments that record the sweetness and sadness of Bao Ma, but also describe the "happiness of the chopping party", each of which is unique, fully demonstrating the infinite fun and potential of "Tiangong SkyMusic" in the field of music creation.
Fang Han believes that traditional music creation often requires a deep professional background and high production costs, but "Tiangong SkyMusic" uses AI technology to make it easy for more people to participate in music creation, whether it is the conception of melody, the arrangement of harmony, or the writing of lyrics, all can be assisted by "Tiangong SkyMusic", which greatly broadens the scope of participation in music creation.
Whether it is the high-quality music output or the ability to accurately control the emotional color of the music, it demonstrates the leading position of "Tiangong SkyMusic" in the AI music model industry. This is inseparable from the strong technical support behind it - the "Tiangong 3.0" pedestal model.
"Tiangong 3.0" has 400 billion parameters, surpassing the 314 billion parameters of Grok-1, and is one of the world's largest open-source MoE models. "Tiangong 3.0" has breakthrough performance improvements in the fields of semantic understanding, logical reasoning, generalization, generalization, uncertainty knowledge, and learning ability, and the ability of mathematics/reasoning/code/cultural creativity has been improved by more than 30%.
Based on the powerful capabilities of "Tiangong 3.0", "Tiangong SkyMusic" has been able to demonstrate its excellent music generation technology and bring users an unprecedented music experience.
It is worth mentioning that "Tiangong SkyMusic" not only has significant advantages in terms of technology, but also shows a profound imagination in the social value of "cultural equality".
Fang Han pointed out that music, as a form of cultural expression, can often transcend the boundaries of geography and language, and become a bridge for communication between different cultures. By lowering the threshold and cost of music creation, "Tiangong SkyMusic" enables more people from different backgrounds and cultures to express themselves through music, thereby promoting cultural diversity and exchange, and to a certain extent, it is conducive to the realization of "cultural equality". Through "Tiangong SkyMusic", everyone can use music to express themselves to their heart's content.
Looking back on the rise of "Tiangong SkyMusic", we can't help but admire the efforts of the technical team behind it, who not only brought excellent performance to "Tiangong SkyMusic", but also injected new vitality into the entire AI music model industry.
It is believed that with the continuous optimization of "Tiangong SkyMusic" in the future, Kunlun Wanwei will continue to play a leading role in the field of AI music, promote the diversity of music creation and the development of cultural exchanges, and bring the fun and infinite possibilities of music creation to more people!