Google to Create Accurate Online Speaker Diarization Tool

The development is based on a recurrent neural network
14 November 2018   745

Google reported on the creation of an innovative diarization algorithm - dividing the incoming audio stream into homogeneous segments in accordance with the belonging of words to a particular person. The company claims that the technology created is more efficient than previously known.

The development is based on a recurrent neural network (RNN). This architecture allows the use of internal memory for processing sequences of arbitrary length and is well suited for working with split audio. In the development of Google for each speaker stands out a separate copy of the RNN, isolating the statements.

Google experts note that their algorithm is completely transparent and controllable, which allows you to adjust the processing of the audio stream.

The developers tested the effectiveness of the new diarization algorithm using the NIST SRE 2000 CALLHOME test. The determination error was 7.6%. The previously used methods of clustering and selection using a neural network showed an error of 8.8% and 9.9%, respectively. In addition to fewer errors, the algorithm has sufficient performance to process the stream in real time.

The definition of replica ownership is an important component of the speech recognition system. Correct diarization allows to adapt better to the peculiarities of pronunciation and accent and to qualitatively separate the statements of different people. The technology will be used, in particular, in creating subtitles for video recordings. Properly recognized speech is easier to translate into other languages, which, for example, would be useful for online training courses. And the ability to process sound in real time will allow you to do it even live.

Nvidia to Open SPADE Source Code

SPADE machine learning system creates realistic landscapes based on rough human sketches
15 April 2019   634

NVIDIA has released the source code for the SPADE machine learning system (GauGAN), which allows for the synthesis of realistic landscapes based on rough sketches, as well as training models associated with the project. The system was demonstrated in March at the GTC 2019 conference, but the code was published only yesterday. The developments are open under the non-free license CC BY-NC-SA 4.0 (Creative Commons Attribution-NonCommercial-ShareAlike 4.0), allowing use only for non-commercial purposes. The code is written in Python using the PyTorch framework.

Sketches are drawn up in the form of a segmented map that determines the placement of exemplary objects on the scene. The nature of the generated objects is set using color labels. For example, a blue fill turns into sky, blue into water, dark green into trees, light green into grass, light brown into stones, dark brown into mountains, gray into snow, a brown line into a road, and a blue line into the river. Additionally, based on the choice of reference images, the overall style of the composition and the time of day are determined. The proposed tool for creating virtual worlds can be useful to a wide range of specialists, from architects and urban planners to game developers and landscape designers.

Objects are synthesized by a generative-adversarial neural network (GAN), which, based on a schematic segmented map, creates realistic images by borrowing parts from a model previously trained on several million photographs. In contrast to the previously developed systems of image synthesis, the proposed method is based on the use of adaptive spatial transformation followed by transformation based on machine learning. Processing a segmented map instead of semantic markup allows you to achieve an exact match of the result and control the style.

To achieve realism, two competing neural networks are used: the generator and the discriminator (Discriminator). The generator generates images based on mixing elements of real photos, and the discriminator identifies possible deviations from real images. As a result, a feedback is formed, on the basis of which the generator begins to assemble more and more qualitative samples, until the discriminator ceases to distinguish them from the real ones.