SoundHound’s Automatic Speech Recognition (ASR) team is currently developing and starting to roll out a state-of-the-art speech recognition system and model architecture improve accuracy, reduce complexity and cost, and enable new dynamic features to make the system more scalable and easily customized. This project (known as Polaris) requires many cross-functional disciplines and close collaboration to be successful. Our team (ASR Systems) is responsible for the most intensive software engineering tasks relating to developing Polaris and supporting our existing speech recognition system in production. This particular role is for our Inference Squad, which is a highly specialized group primarily responsible for high-performance inference and model integration. Members of the Inference Squad provide the necessary expertise to ensure alignment between the systems-focused engineers and the ASR Modeling team who focus on research and model development.

In this role, you will:

• Enhance the core of our proprietary inference server (Snowcat) to support Polaris rollout and integration with various products
• Help support our existing production ASR system until we successfully replace everything with Polaris, including targeted enhancements, incident investigations, and technical analysis as needed

We would love to hear from you if:

Required:

• You have excellent general software engineering skills and discipline
• You have professional experience with both C++ and Python for complex systems
• You have exceptional attention to detail when solving hard technical problems, including design, refactoring, optimization, and debugging
• You are thoughtful in decision making, especially in understanding business priorities, balancing trade-offs, and managing ambiguities
• You have advanced expertise with algorithms and data structures

Nice-to-Have:

• You have a strong math background (e.g. digital signal processing, machine learning, control theory, geometry, or probability theory)
• You have expertise with low-level C++ concepts such as memory management and bit-level arithmetic
• You have a deep understanding of programming with tensor math libraries (e.g. NumPy, PyTorch, Eigen)
• You are familiar with CUDA programming and GPU architecture
• You have an understanding of advanced indexing used in tensor math libraries
• You have experience with TorchScript
• You have experience with finite state transducers, abstract syntax trees, hidden Markov models, and beam search

We’d be especially excited if:

• You have specific experience with speech recognition or general interest in linguistics.

This role is available throughout Canada. Employees within a 100-kilometer radius of our Toronto office are expected to work from the office on three pre-scheduled “core days” each month to encourage cross-team connection and in-person collaboration. Aside from these office-specific “core days,” this job allows for virtual/remote, hybrid, and in-office workplace setting options. In addition to salary and equity, you will receive comprehensive healthcare, paid time off, and other benefits. Our recruiting team will provide a specific salary range based on location and years of experience.

_______________________

#LI-MR1

#LI-HYBRID