Zyphra
Research Engineer - Audio & Speech Models
San Francisco · On-site · Full-time · Posted Mar 17, 2026
- Education
- Postgraduate degree in a scientific subject (Computer Science, EE/EECS, Mathematics, Physics, Machine Learning)
- Visa sponsorship
- Relocation and immigration support on a case-by-case basis.
About the job
Build and train open-source audio and speech models, working across data, model architecture, training methods, and performance optimization.
What you'll do
- Run large-scale audio model training and optimize the training stack.
- Collect, process, and evaluate audio datasets.
- Develop and test model architectures and training methodologies through ablations and improvements.
What they ask for
- Strong research judgment and ability to carry a project from conception through execution and write-up.
- Strong implementation and prototyping ability.
- Ability to collaborate effectively in a fast-paced research setting.
- Excellent communication and collaboration skills across research and engineering work at scale.
Nice to have
- Audio-model training experience, including autoencoders; signal-processing knowledge; experience with diffusion models, consistency models, or GANs; and large-scale multi-node GPU training.
- PyTorch and Python proficiency; experience with large codebases, rigorous experimental methodology, or parallel data pipelines; published ML research; postgraduate scientific degree.
Tools
PyTorch, Python
AI focus
Research and engineering of open-source audio and speech models, including autoencoders, ASR, TTS, SSL, and speech-to-speech models.
Model-written summary · the original posting is authoritative