Research Staff, Voice AI Foundations

About the role

As a member of the research staff, you will pioneer the development of Latent Space Models to address fundamental data, scale, and cost challenges in voice AI. You will focus on building next-generation neural audio codecs, steerable generative models, and embedding systems to enable robust, contextualized voice interaction. This role requires a deep commitment to experimental rigor, algorithmic innovation, and the ability to bridge theoretical research with efficient, scalable implementation.

Responsibilities

• Build next-generation neural audio codecs for extreme, low bit-rate compression and high fidelity reconstruction.
• Pioneer steerable generative models capable of synthesizing diverse human speech from codec latent representations.
• Develop embedding systems that factorize latent space into interpretable dimensions for precise control over speaker, style, and environment.
• Leverage latent recombination to generate synthetic audio data at scale for model training.
• Design model architectures, training schemes, and inference algorithms optimized for bare-metal hardware efficiency.

Required skills

• Statistical learning theory
• Foundation model architectures
• Multimodal learning
• Neural audio codecs
• Generative models
• Data pipeline development
• Experimental design
• Model optimization

Qualifications

• Strong mathematical foundation in statistical learning theory
• Deep expertise in foundation model architectures
• Proven ability to bridge theory and practice
• Track record of open-source contributions or research publications in speech/language AI

About the Company

Deepgram is a platform for the Voice AI economy, providing real-time APIs for speech-to-text and text-to-speech, and building production-grade voice agents at scale. The company's voice-native foundation models are used by over 200,000 developers and 1,300 organizations to build voice offerings.