Senior Machine Learning Scientist (ASR)

Who you are

• 8+ years experience in machine learning research & engineering for voice applications
• A PhD or Master’s degree in computer science, or a related technical field (or equivalent experience)
• In-depth knowledge of speech processing (ASR, wakeword detection, audio features), particularly latest architectures for automatic speech recognition (RNN-T, Transformers)
• Experience in owning the whole model development lifecycle from data generation, training, evaluation to shipping models for production
• Advanced knowledge of Python and common machine learning toolkits (PyTorch)
• Intermediate knowledge of a low-level compiled language (Rust, C, C++)
• Preferred Skills
• Experience in developing models for real-time, low latency, streaming ASR
• Experience in model adaptation techniques for large entity catalogs (shallow-fusion biasing, rescoring, error correction)
• Understanding of the particular challenges of far-field ASR, including acoustics and data augmentation for model training and evaluation
• MLOps and software engineering for efficient training and data infrastructure

What the job involves

• We are seeking an experienced ML engineer to join the Audio Machine Learning team, focused on building the next generation of Sonos Voice Control.
• The team develops all audio-based components of Sonos' in-house voice assistant solution, including far-field automatic speech recognition (ASR), wakeword detection, and speech enhancement.
• The team is also in charge of shipping those models to production and running them efficiently and fast on multiple types of hardware
• Work in the Audio Machine Learning team of Sonos Voice Control, together with a group of experienced machine learning engineers and researchers
• Design and train state-of-the-art machine learning models for automatic speech recognition and wakeword detection
• Ensure our models perform the best they can on specialised domains, such as music entities, loudspeaker control and home automation
• Define and implement the data strategy for training and testing, as well as audio augmentation to reflect the far-field acoustic conditions of our products
• Maintain and improve scalable and efficient training and evaluation pipelines
• Contribute to the team’s roadmap and set the technical direction for the ASR domain
• Mentor other team members on ML best practices, experiment planning, and model analysis
• Collaborate with the cloud backend and embedded engineering teams to ensure our models perform the best they can in the different environments

The application process

• Your profile will be reviewed and you'll hear from us once we have an update. At Sonos we take the time to hire right and appreciate your patience