Advance Search

Browse CVs

Machine Learning Engineer

Posted 4 months ago

About You You are an experienced Machine Learning Engineer who can help us advance our automatic speech recognition (ASR) and is excited to build the voice interfaces of the future. We are betting big on scaling models so you will be working with millions of hours of audio and billion parameter models which train across dozens of GPUs. It s all about finding the bottlenecks across our 30+ languages and targeting our efforts to understand every voice . Our main research focus is large-scale Self-supervised Learning and a practical focus on building state-of-the-art speech pipelines. An average day might include working on Scaling self-supervised learning models across hundreds of GPUs in the cloud Experimenting with distillation or quantisation to speed up our models at runtime Comparing compute efficiencies of architectures such as a transformer and the impact on WER Developing a new product, such as Language ID, from training the model all the way to shipping it to production Advancing end-to-end speech models in PyTorch such as the RNN Transducer Giving a journal club on a recent paper such as DALLE-2 We aim to get you onboarded and started on something like this in your first few days. In addition, having a very collaborative culture, you will often be pair programming with a colleague on streamlining our production ML pipelines, reviewing other folks code and suggesting new ways to tackle a tough real time factor (RTF) optimisation problem as well as brainstorming novel approaches for analysing model predictions with the team. You ll want to join our team if you Want to learn more about speech recognition and representational learning Are results driven, like moving fast and keeping things simple Love working in collaborative and ambitious teams Have a growth mindset and love to develop yourself and others Enjoy solving challenging problems and digging into a stack of unfamiliar code Love optimising code You may have experience in some of the following A GitHub portfolio demonstrating personal ML projects Large scale distributed model training Deep learning and Pytorch Language modelling and acoustic modelling Comfortable on the command line and shell scripting ETL pipelines in Python (we like Prefect and Airflow) Speaking multiple languages or a background in linguistics About Us Speechmatics has collaborative office spaces in the UK, with teams also located in the Czech Republic, India and the US. Our speech-to-text software is the world s most accurate, regularly winning customers from Google, Amazon and Microsoft (learn more in our whitepapers here ). We believe we are the UK s most exciting ML start-up and have grand ambitions for the future of speech tech. Come and join our mission to build perfect speech recognition to understand every voice! #J-18808-Ljbffr