正在加载视频...
视频加载失败
Transformers can map one mp3 to another mp3. But why do we want to do audio-space translation? SingSong from Google Magenta Project is a super clever application: singing voice -> instrument accompaniments, i.e. “Reverse Karaoke”! Now everyone’s got their private, deluxe band 🎤 1/🧵
10 条评论

Overall idea is simple. Use an off-the-shelf source separation algorithm to make a synthetic training dataset from a large corpus of music audio. Each data point is an aligned (vocal, instrument) pair. Then train a Transformer to predict instrumental sound from vocals. 2/

Inference time is even simpler: directly take input vocals from users and generate the conditional instrument accompaniment. Just naively mix with the input, and voila! It will sound coherent. 3/

Paper: Demos: Original tweet from @chrisdonahuey, congrats to the Magenta team! 4/

“Reverse Karaoke” is a really witty & accurate characterization, quoted from @jesseengel. Welcome to follow me for the latest AI 🙌 END/🧵

@GoogleMagenta The French singing at the end is hilarious.

@GoogleMagenta « Et ça fait Hé, allez les Bleus, on va gagner le foot, à l’aide des pieds, ça va être super bien. On va marquer plus de points que l’adversaireuuuu… Let’s go! » 🤣🤣🤣🤣🤣🤣

@GoogleMagenta It's a dream app for all Asian PhDs

@GoogleMagenta 😜

@GoogleMagenta Le français à la fin! 🤣🤣🤣🤣🤣

@GoogleMagenta Impromptu @kronomuzik à 1:20
