SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Deepgram is the leading platform for the Voice AI economy, providing real-time APIs for speech-to-text, text-to-speech, and production-grade voice agents at scale. The company has processed over 50,000 years of audio and transcribed more than 1 trillion words, serving 200,000+ developers and 1,300+ organizations including Twilio, Cloudflare, and Sierra.
As a Member of the Research Staff, you will pioneer the development of Latent Space Models (LSMs), a new approach to solve fundamental data, scale, and cost challenges in building robust, contextualized voice AI. Your research will focus on one or more of these core problems:
• Build next-generation neural audio codecs achieving extreme, low bit-rate compression and high-fidelity reconstruction across world-scale audio corpora.
• Pioneer steerable generative models synthesizing the full diversity of human speech from codec latent representations, including casual conversation, emotional expression, multi-speaker scenarios with environmental noise, and overlapping speech.
• Develop embedding systems that factorize codec latent space into interpretable dimensions (speaker, content, style, environment, channel effects), enabling precise control and dataset amplification through "latent recombination."
• Leverage latent recombination to generate synthetic audio at unprecedented scales, unlocking new joint model and data scaling paradigms for audio.
• Train multimodal speech-to-speech systems that understand any human regardless of demographics, state, or environment, and produce empathic, human-like responses for conversational or task-oriented objectives.
• Design model architectures, training schemes, and inference algorithms optimized for hardware at bare metal, enabling cost-efficient training on billion-hour datasets and real-time inference for hundreds of millions of concurrent conversations.
Deepgram operates with an AI-first mindset where all team members actively use and experiment with advanced AI tools. The company moves at the pace of AI, with rapid change and continuous evolution of day-to-day work. Success requires comfort with experimentation, adaptation, and constant learning.
You should see unsolved problems as opportunities to pioneer entirely new approaches, identify critical experiments that validate or kill ideas in days rather than months, and have the vision to scale successful proofs-of-concept 100x. The role demands obsession with problems, creativity in approach, and relentless drive toward elegant, scalable solutions.
REQUIREMENTS:
• Strong mathematical foundation in statistical learning theory, particularly in self-supervised and multimodal learning
• Deep expertise in foundation model architectures and understanding how to scale training across multiple modalities
• Proven ability to bridge theory and practice—deriving novel mathematical formulations and implementing them efficiently
• Demonstrated ability to build data pipelines processing and curating massive datasets while maintaining quality and diversity
• Track record of designing controlled experiments isolating the impact of architectural innovations and validating theoretical insights
• Experience optimizing models for real-world deployment, including knowledge of hardware constraints and efficiency techniques
• History of open-source contributions or research publications advancing the state of the art in speech/language AI
The ideal researcher excels in five key areas: (1) Statistical & Mathematical Foundations—mastery of core concepts, combining existing theories in novel ways, and logical reasoning with assumption testing; (2) Algorithmic Innovation & Implementation—creative solutions to known bottlenecks and turning theory into practice; (3) Data-Driven & Scalable Systems—emphasis on large-scale data, efficient pipelines, and strategic use of data for staged training; (4) Hardware & Systems Understanding—efficient implementations at scale, real-time and on-device constraints, and architectural optimization details; (5) Rigorous Experimental Design—controlled comparisons with ablations, multifold evaluation metrics, and stress tests on edge cases.
