Machine Learning Scientist, Abridge

• We're currently looking for a research scientist with experience in speech recognition to help us improve the performance and robustness of our transcription modules
• The ideal candidate will bring technical mastery, fluency with foundation models, speech recognition, genuine interest in the medical domain, and strong critical thinking skills to the role
• 2+ years experience in building speech recognition systems
• Strong research background, as demonstrated through papers and a graduate degree (MS or PhD) in Electrical Engineering, Computer Sciences, Mathematics, or equivalent experience with a specialization in speech recognition or machine learning
• High-impact publications at peer-reviewed speech conferences (e.g. ICASSP, Interspeech, SLT) or NLP/ML conferences (NAACL, ACL, NeurIPS, ICML, ICLR)
• Significant real-world impact, as demonstrated through open-source contributions and deployed technology
• Strong programming skills with proven experience crafting, prototyping, and delivering machine learning solutions into production
• Experience with deep learning libraries (e.g. PyTorch, Jax, Tensorflow) and platforms, multi-GPU training, and statistical analyses of observational and experimental data
• Deep knowledge of ASR technologies such as acoustic modeling, language modeling, neural networks, HMM, WFST, and feature extraction
• Hands-on experience in ASR tool kits such as Sphinx, Kaldi, HTK, or Julius

Desirable

• At Abridge, all of our ML work has a strong research component, and all of our research scientists contribute directly to real products that impact the lives of doctors
• Advance the state of the art in medical ASR, in areas including accurate transcription of clinical conversations, support of a wide variety of languages and dialects, speaker diarization, domain-specific language modeling, robust handling of diverse accents and noise conditions, and development of novel evaluation and experimentation techniques
• Actively contribute to the wider research community by sharing and publishing original research
• Research and implement algorithms and methods for long-form and real-time speech recognition
• Perform data preprocessing and define performance measures based on development and test sets
• Implement and improve tools, algorithms, and prototypes
• Be responsible for measuring and optimizing the quality of your algorithms
• Help to define important problems, identify appropriate baselines, develop state-of-the-art methods, and ship them into production
• Dial deeply into real-time feedback from clinicians to guide further refinements and innovations
• Be results-oriented in the face of ambiguous problems and uncertain outcomes