PixelPlayer to Learn How Extract Musical Instrument

Massachusetts Institute of Technology scientists created new neural network
06 July 2018   652

Scientist from MIT managed to create a neural network called PixelPlayer, which is able to indetify and extact the sound of individual musical instruments. The key feature of the development is the use of the method of spontaneous learning. This is reported by Analytics Vidhya.

In similar developments, the method of controlled learning was previously used. As input data, the AI ​​received marked audio files, the manual marking of which required a lot of time. PixelPlayer processes video - this allows to opt out of the preliminary preparation of information. Spontaneous training eliminated the human factor and accelerated the process.

Development involves three algorithms at once. The first processes the video, the second - the audio track, and the third synchronizes the data. PixelPlayer determines the sound pertaining to each pixel in the image. In this way, the neural network detects individual instruments and determines the melody to be released.

After 60 hours of training, the AI ​​was able to recognize with high accuracy individual melodies on new video recordings that had not been shown to it before. According to the developers, PixelPlayer is able to identify up to 20 different tools. This number can be increased by providing additional data for processing. Errors occur about trying to divide class-like instruments, for example, saxophone-alto and tenor.

PixelPlayer has already considerable potential for practical application. With this tool the quality of old live recordings can be improved. Amateur musicians often try to "remove" a certain party aurally, and the development of MIT scientists can simplify this task.

Nvidia to Open StyleGan Source Code

This machine learning project allows to create of people faces by imitating photographs
11 February 2019   315

NVIDIA has open source code if developments related to the StyleGAN project, which allows generating images of new faces of people by imitating photographs. The system automatically takes into account aspects of the placement of individuals and makes the result indistinguishable from real photos (most of the respondents could not distinguish the original photos from the generated ones). For the synthesis of individuals, a machine learning system based on a generative-competitive neural network (GAN) is used. The code is written in Python using the TensorFlow framework and published under the Creative Commons BY-NC 4.0 license (for non-commercial use only).

Both ready-made trained models and collections of images for self-learning of a neural network are available for download. The basic model was trained on the basis of the Flickr-Faces-HQ (FFHQ) collection, which includes 70,000 high-quality (1024x1024) PNG images of people's faces. At the same time, the system is not tied to persons - as an example, the variants trained on collections of photographs of cars, cats and beds are shown. It requires one or more NVIDIA graphics cards (Tesla V100 GPU recommended), at least 11 GB of RAM, NVIDIA 391.35+ drivers, CUDA 9.0+ tools and the cuDNN 7.3.1 library.

The system allows you to synthesize the image of a new face based on interpolation of features of several faces, combining features characteristic of them, as well as adapting the final image to the required age, gender, hair length, smile character, nose shape, skin color, glasses, face rotation in the photo. The generator considers the image as a collection of styles, automatically separates the characteristic details (freckles, hair, glasses) from common high-level attributes (posture, gender, age changes) and allows you to combine them in an arbitrary form with the definition of the dominant properties through weights.