Scientists from MIT have created a neural network PixelPlayer that able to distinguish musical instruments in the video. With its help it is possible to selectively disable the tools and make them louder, according to Engadget.
PixelPlayer unique in that it works directly with the three algorithms. The neural network determines which tool is depicted in the video, then allocates it to an audio track, then sinhroniziruete data. To teach her this, scientists used a variety of music videos. Their total duration exceeds 60 hours.
After training, the neural network learned to distinguish between musical instruments that are depicted in the video. According to the developers, PixelPlayer able to identify up to 20 different tools. Engadget notes that the algorithm still works not perfect. For example, it may be wrong in the interaction with similar tools.
Development at MIT will come in handy when editing old audio or concert audio recordings. PixelPlayer can also be used to teach robots the means of identifying different environmental sounds such as animals, vehicles and appliances.
MDlavka — our store for your favorite readers. Technology Apple for the most pleasant prices are waiting for you every day from 10:00 to 21:00. Macdigger putting the promo code at checkout and get discount of 500 RUB.
Follow the news in our Telegram-channel, as well as in the app on iOS MacDigger.