Jiemian News Reporter | Yu Hao Jiemian News Editor | Wen Shuqi inputs the music genre, content theme, and can generate 2 minutes of music in just a few seconds... Generative artificial intelligence once again brings people in the field of music Come and be surprised. Recently, ar

interface news reporter | Yu Hao

interface news editor | Wen Shuqi

Enter the music genre, content theme, and generate 2 minutes of music in just a few seconds... Generative artificial intelligence is again in the field of music Surprise people.

Recently, the artificial intelligence start-up suno released the v3 version of the suno model. According to reports, based on the v2 version, the v3 version adds more music styles and genres, while also strengthening compliance with prompt words and reducing hallucination problems.

Suno v3 currently supports custom mode. Users can generate music works by inputting lyrics, titles, music styles, and checking whether it is purely instrumental music in the custom mode. They can also use simpler prompt words in non-custom mode. Generate works.

interface news uses "a jazz song that describes happy love (a jazz song that describes happy love)" as the prompt word to enter suno v3. The following is the link to the music work and lyrics generated by the model: (lyrics translation is completed by kimi intelligent assistant)

https://app.suno.ai/song/c066c053-e038-493b-a48d-095a9891b1cb

the moon is shining bright above

the moon is shining bright above

as we sway to the rhythm of love

we sway gently to the rhythm of love

your touch, it sets my heart on fire

Your touch is like fire igniting my heart

in this moment, nothing could be higher

the sweet melody fills the air

as we dance without a care

We dance carefree

your eyes, they sparkle like stars

in this jazz embrace, we're taking it far

in the embrace of jazz, we travel far away

oh, this love is like a moonlit serenade

Oh, this scene is like a serenade under the moon

a harmony that will never fade

harmony melody, never fade

in your arms, i find my solace and peace

in your arms, I find peace and tranquility

together, we create a love masterpiece

work together to draw, love's masterpiece

According to Rolling Stone magazine, the suno entrepreneurial team was established less than two years ago and was co-founded by Mikey Shulman, Keenan Freyberg, Georg Kucsko and Martin Camacho. The four are all experts in machine learning and have previously worked together at the artificial intelligence company kensho. Among them, Shulman and Camacho are also musicians and often perform impromptu performances together. In terms of technical path,

is similar to large language models such as chatgpt. Suno decomposes human language into token units, absorbs its usage and structure, and then re-outputs it according to needs. But unlike plain text, which can be easily split, audio often appears in the form of continuous signals, which makes it more difficult for AI to understand and generate music. The similar training ideas of

make suno likely to face the copyright lawsuit suffered by openai. In December 2023, the New York Times sued OpenAI and Microsoft for copyright infringement, requiring the two companies to destroy any chat robot models and training data that used copyrighted materials of The New York Times. The founding team of suno did not disclose the training data of the model, only that the model will learn the characteristics of human voices by learning voice recordings. After the release of

suno v3, the suno founding team posted on social media that the suno model is still in its early stages and will be improved in terms of quality, controllability and speed in the future, and revealed that v4 is under development. At the same time, in order to prevent the misuse of model works, suno has also developed proprietary silent watermark technology to detect whether the song was created using suno.

In terms of artificial intelligence music creation, Google and stability ai have released similar products. Google's dream track allows users to use celebrity voices to create their own songs, and stable audio ai released by stability ai can generate a track of about 45 seconds. .

Suno currently only has about 12 employees, but in the above report the founding team stated that they plan to expand. At the same time, they also said that the current suno model is not close to perfection, and there is still a lot of work to be done.One direction is that suno hopes to replace the text-to-music solution with a more intuitive interactive method, such as generating songs based on the user's own singing.