The voice assistant Alice can distinguish user commands even in strong noise and loud music. This is made possible by noise and echo suppression algorithms developed by our team. Additionally, our algorithms process sound during calls via smart devices and the Telemost service.

We are actively developing neural network approaches and are looking for an experienced ML engineer who can strengthen our team. We value experience in ML, and we are ready to teach the specifics of working with sound.

For our work, we have:

• hundreds of modern GPUs for experiments and colossal volumes of data;
• a platform for A/B testing and data labeling with support from professional analysts;
• acoustic laboratories and an apartment where we record sound.

What tasks await you

Improving the quality of noise and echo suppression models You will experiment with architectures, data, and training schemes to make our models even better.

Implementation and optimization of models for embedded devices Our models run on Yandex devices with Alice, as well as on users' smartphones and laptops. Operational efficiency is a key aspect. You will be responsible for speeding up models at both the level of architecture/parameter selection and final implementation.

Creating and improving quality metrics We are developing metrics that will be computed on users' devices while preserving their privacy and will serve as an indicator of overall sound quality.

We expect you to have

• Worked in the field of DL, for example with CV or NLP
• Be well-versed in Python and have basic C++ skills
• Know PyTorch

Will be a plus

• Have deployed neural networks to production, especially on user devices
• Have advanced C++ skills
• Have worked with large volumes of data
• Are familiar with Yandex's internal infrastructure or cloud computing platforms: AWS, Azure, Google Cloud
• Have experience in web development
• Have worked on classical digital signal processing