Wearing a cap, sunglasses and a big gold chain, lu Yang, a programmer born in the 1990s who has never played music and is even tone-deaf, seems like a very distant dream.
But recently, he did record his first rap with a friend on Ali Radio, about Double 11 and artificial intelligence. Helping him do this was MusicGo, an artificial intelligence he designed himself.
MusicGo, an artificial intelligence that can write its own rap, can come up with rhyming FreeStyle anytime, anywhere, just give it a key word.
Before this summer, Lu Yang knew almost nothing about rap. He was tone-deaf and fell in love with hip-hop purely by accident.
Lu Yang, who usually pays little attention to variety shows, went to the show in order to “chat with colleagues”, but unexpectedly fell in love with it. “I’m based in Beijing, but I’ve been to Hangzhou a lot this year on business. I spent a lot of time on the road, so I watched the program while on business trips and on my daily commute. The more I watched it, the more I liked it and watched it several times.” Lu Yang said that while watching the program, he also understood flow, punchline, as well as the cool double, triple and many other professional words.
Lu Yang is a big fan of the Rap of China, an alternative form of music that appeals to him, as well as the singers’ free and easy attitude and their repeated “Keep Real” spirit
The more I watched, the restless programmer gene in Luyang began to be agitated, and the desire to analyze rap data grew stronger. “I was particularly interested in rhyme. I was thinking, wouldn’t it be cool if I could take down all the rhyming words that hip-hop singers sing and just give them a key word and FreeStyle?”
So an AI he wrote in his spare time called MusicGo tried to get MusicGo to write its own rap:
He found playlists on the Internet with the “Hip-hop” label, pulled all the songs from the playlist, and then removed the duplicate songs. Then “crawl” the song page of the song list in turn to get all the detailed song list. In total, more than 10,000 songs were retrieved. Then, over the course of a few hours, we get over 10,000 lyrics by calling the lyrics API, based on the song list.
Next, Lu Yang uses an algorithm to divide songs into words, such as “I love Beijing Tian ‘anmen” into “I love”, “Beijing” and “Tian ‘anmen”, and then separates each group of pinyin, such as “Tian ‘anmen” into “Tian-An-men”.
“So we have the words of all the songs, the frequency of the words (word frequency), and the pinyin corresponding to the words. Once you have pinyin, you have to consider rhymes.” After observation, Luyang found that “A-e-I-O-U” could be used as the boundary of the character, and the current and the following part could be used as the rhyme.
Using PG One’s favorite “chow mein” FreeStyle from the Rap of China, Lu tried the rhyming of “fried rice” and a series of related terms emerged. “These words all rhyme with fried rice, and I tried to write a rap with these words, and in 20 minutes I wrote a rap with 19 double strokes.”
The article, which was originally posted on a tech forum, was quickly posted on its first screen and received more than 200 likes and 1,000 views within the day.
“Feed” intelligent articles “spit” out of the sense of technology lyrics
But getting MusicGo to use an algorithm to write lyrics that rhyme with the theme and then make the substitutions itself is semi-automatic, not luyang’s ultimate goal.
“The theme of This year’s Singles’ Day is artificial intelligence, which is more about cooperation between humans and machines. I thought about it and I thought it was pretty cool. Cool is the first productive force. I just want ai to write a rap about Double 11.” LuYang said.
Lu Yang joined the rookie only four months, this is his first time to participate in double 11 preparation
So he revamped MusicGo. What makes MusicGo unique is that instead of relying on developers to manually input hundreds of interval relationships and rhythmic patterns, its core skill is “machine learning.”
By feeding thousands of completed songs into the system, MusicGo ‘learns’ the patterns and patterns of rap on its own, creating a’ neural network ‘that can be composed.
When Luyang typed keywords through a computer keyboard, it looked for and recognized similar lyrics, melodies and rhythms in a network of neurons.
The essence of machine learning is that it understands musical concepts such as “rhythm” and “tonal consistency” from a statistical perspective, rather than from a sound perspective.
“After ‘learning’ a lyric, it can ‘recreate’ another lyric, so I just need to adjust a few rhymes and the whole song is relatively complete. “Said Lu Yang.
MusicGo uses machine learning technology, which means that the machine automatically learns whatever content is “fed” to it by a human. To produce a double eleven and ai song, it is necessary to “feed” a lot of science and technology content, so Luyang searched hundreds of online articles related to double Eleven, intelligence, Ali Cloud, cainiao and so on, and fed them to MusicGo. And it worked. MusicGo spat out tech-savvy lyrics.
Therefore, he set the rhythm and mode he wanted first, and then input the theme of “Double 11”. According to the same routine previously, a section of lyrics was generated in this way “second level”. Lu Yang also used 20 minutes to modify the rhyme, completed the first “Tmall double 11 hip-hop” RAP, Lu Yang put the lyrics on the Intranet, immediately like-minded colleagues came to the chorus.
Tmall Double 11 Hip Hop
Lu said that while MusicGo is not yet fully intelligent, ai can do more and more things. In fact, Aliyun’s AI technologies, including speech recognition, image recognition, face recognition, speech synthesis, natural language understanding and machine learning, have been applied to traffic prediction, intelligent customer service, court shorthand, weather prediction and other fields.
“Now I can write lyrics, and in the future I might write a girl ticket.” The big boy had one last bit of humor.
The original post was published on November 9, 2017
Author: Wang Jiajing
This article is from the cloud community partner “Tianxia Network”. For relevant information, you can follow the wechat public account of “Tianxia Network”