Member of Research Staff (AI)

Location: Remote, London, UK

Company

Deepgram provides real-time voice AI APIs and self-hosted software for speech-to-text, text-to-speech, and production voice agents.

What you will do

• Develop latent space models addressing the data, scale, and cost challenges of robust contextualized voice AI.
• Build neural audio codecs with extremely low bit rates and high-fidelity reconstruction across diverse audio.
• Design steerable generative models for expressive, multi-speaker speech with environmental noise and overlapping speech.
• Create embedding systems that factorize speaker, content, style, environment, and channel effects.
• Generate synthetic audio data through latent recombination and develop multimodal speech-to-speech systems.
• Optimize architectures, training methods, and inference algorithms for efficient large-scale training and real-time deployment.

Requirements

• Strong mathematical foundation in statistical learning theory, especially self-supervised and multimodal learning.
• Deep expertise in foundation model architectures and scaling training across multiple modalities.
• Ability to derive novel mathematical formulations and implement them efficiently.
• Experience building and curating massive, high-quality data pipelines.
• Experience designing controlled experiments, ablations, and rigorous evaluation methods.
• Knowledge of hardware constraints, model efficiency, speech or language AI, and real-world deployment.

Nice to have

• Open-source contributions or research publications advancing speech or language AI.
• Experience with neural audio codecs, generative models, quantization, efficient architectures, or custom hardware kernels.

Culture & Benefits

• AI-first working environment with an expectation of actively using and experimenting with advanced AI tools.
• Fast-changing research environment focused on novel approaches to unsolved problems.
• Remote work arrangement associated with London, UK.
• Emphasis on scalable solutions, rigorous experimentation, and impactful open research.