Senior Machine Learning Scientist (ASR)

Job description

At Sonos we want to create the ultimate listening experience for our customers and know that it starts by listening to each other. As part of the Sonos team, you’ll collaborate with people of all styles, skill sets, and backgrounds to realize our vision while fostering a community where everyone feels included and empowered to do the best work of their lives. This role is a hybrid position This position is considered hybrid, allowing for a combination of remote work and in-office collaboration. Qualified applicants must live within commuting distance of our Paris office location and should expect to be in office approximately 3 days per week. In the Sonos Voice Control team we design the future of AI based interactions to power music control and content discovery for Sonos customers on any control surfaces (Sonos hardware, Sonos Application, Sonos Voice Control). 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. What You’ll Do 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 What You’ll Need Basic Qualifications 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 Research shows that candidates from underrepresented backgrounds often don't apply for roles if they don't meet all the criteria. If you don’t have 100% of the skills listed, we strongly encourage you to apply if interested. #LI-hybrid 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.