Member of Technical Staff, Applied AI – Speech

Role overview

Build and improve real-time speech technology for production voice agents handling customer servicing conversations. You’ll own important parts of the speech stack, taking problems from investigation through deployment and improving latency, conversational flow, naturalness, and reliability under real call conditions.

Responsibilities

- Design, train, and ship speech-to-speech and real-time voice models, improving turn-taking, interruption handling, robustness, and conversational quality. - Develop production components such as streaming inference, neural codecs, speech recognition or synthesis pathways, turn detection, and speech enhancement. - Create evaluation tools and monitoring based on real conversations, so performance reflects live conditions rather than clean benchmarks alone. - Investigate issues caused by telephony noise, accents, channel differences, overlapping speech, and callers interrupting the agent; turn findings into improvements to models, data, or systems. - Work across models, data, infrastructure, and product with applied AI and voice teams to improve end-to-end agent performance. - Take technical areas from problem definition through production release with substantial autonomy.

Requirements

- At least 5 years of experience in speech, audio, or related machine learning, or equivalent demonstrated technical depth. - Evidence of building speech-to-speech, full-duplex, or other real-time conversational speech systems; experience limited to offline ASR or TTS is not the stated focus. - Publicly inspectable work, such as papers, code repositories, models, or datasets. - Strong software engineering skills, including experience building reliable, low-latency or streaming systems. - Ability to diagnose real-world model failures and translate them into practical improvements. - Strong ownership and execution in technically ambiguous areas.

Nice to have

- Experience with neural audio codecs, vocoders, or discrete speech representations. - Background in turn detection, barge-in, speech enhancement, multilingual speech, or production telephony and contact-center systems. - Experience adapting language models for voice agents and developing evaluations that predict production quality. - Familiarity with regulated or high-stakes conversations.

Benefits and work setup

- The role is described as fully remote, with close collaboration with an applied AI team based in San Francisco. The listing also includes conflicting details about in-person office work, so the exact arrangement is unclear. - Compensation is described as market-competitive and including equity; benefits depend on the country of hire. Medical, dental, and vision coverage, a retirement plan, and catered lunches are also mentioned.