Research Engineer

Acceler8 Talent; San Francisco (CA); On-site; USD 150,000 - 230,000; Full time

Benefits offered by this job

Equity

Job summary

Acceler8 Talent is seeking a Research Engineer to join a frontier AI research lab focused on real-time conversational voice systems. The role emphasizes building ultra-low latency streaming and full-duplex speech architectures, turning research into production capabilities.

You will prototype speech foundation models, advance cascaded ASR → LLM → TTS pipelines, and help shape evaluation frameworks for conversational quality and social presence.

Qualifications

• Experience researching speech foundation models or voice AI systems.
• Hands-on training, fine-tuning, evaluating, and benchmarking speech models.
• Experience with streaming or low-latency speech systems.
• Exposure to full-duplex conversational models such as Moshi, PersonaPlex, or similar research is desirable.
• Publications at leading speech, audio, or ML conferences are advantageous.
• Strong ML research background with ability to translate research into production.

Responsibilities

• Research and prototype ultra-low latency streaming and full-duplex speech systems.
• Develop speech foundation models, including cascaded (ASR → LLM → TTS) architectures.
• Improve conversational behaviors including turn-taking and interruption handling.
• Train, fine-tune, benchmark, and evaluate state-of-the-art speech models.
• Design evaluation frameworks measuring conversational quality and social presence.
• Collaborate with researchers building multimodal and voice foundation models.

Skills

Speech foundation models

Low-latency streaming

Production-ready ML systems

Publications (ICASP / Interspeech / Ne

Full-duplex models

Research-to-production translation

Tools

Moshi

PersonaPlex

Job description

Research Engineer (Voice / Speech AI)

San Francisco, CA (Onsite)

I am seeking aResearch Engineerto join a frontier AI research lab building the next generation of real-time conversational voice systems. Rather than building traditional AI assistants, this team is creating full-duplex AI companions that can naturally participate in live multiplayer experiences: listening, responding, interrupting appropriately, understanding shared context, and interacting alongside human players in real time.

Alongside building production voice agents, the team also operates as a voice research lab, producing research-grade speech datasets, evaluations, and tooling that support frontier AI research across the wider ecosystem.

What We're Looking For:

• Experience researching speech foundation models or voice AI systems
• Hands-on experience training, fine-tuning, evaluating, and benchmarking speech models
• Experience working on streaming or low-latency speech systems
• Exposure to full-duplex conversational models such asMoshi,PersonaPlex, or similar research is highly desirable
• Publications at leading speech, audio, or machine learning conferences (ICASSP, Interspeech, NeurIPS, ICML, ICLR, etc.) are a significant advantage
• Strong machine learning research background with the ability to translate research into production systems

What You'll Do:

• Research and prototype ultra-low latency streaming and full-duplex speech systems
• Develop speech foundation models, including cascaded (ASR → LLM → TTS) and speech-to-speech architectures
• Improve conversational behaviours including turn-taking, interruption handling, timing, prosody, and emotion
• Train, fine-tune, benchmark, and evaluate state-of-the‑art speech models
• Design evaluation frameworks that measure conversational quality and social presence—not just transcription accuracy
• Work across model research, inference optimisation, data generation, and production deployment
• Collaborate closely with researchers building frontier multimodal and voice foundation models

This is a genuinely research-focused opportunity for someone who wants to push the boundaries of speech AI. You'll be tackling challenges around conversational timing, streaming inference, interruption handling, latency optimisation, and voice quality—building systems that feel socially present rather than simply responsive.

Based indowntown San Francisco, the team works fully onsite and offers highly competitive compensation alongside meaningful equity as they continue to build one of the most ambitious voice AI research groups in the industry.

Ethan Lewis

[e-mail]