Machine Learning Researcher, Audio (TTS) The text-to-speech model already works really well. The job is the last 10% of post-training: the pronunciation errors and small issues that only show up live. About the company A well-funded voice AI company in San Francisco that builds AI phone agents for enterprise customers and now trains its own speech models. The research team has just finished evaluating its newest model and is moving into full development. Research headcount is growing, and text-to-speech is now the main hiring priority. The role You’ll own the text-to-speech model from research prototype to production. That means post-training it until it’s reliable and stable, fixing mispronunciations and the issues that only appear live, and building on neural audio codecs and discrete audio representations. You’ll ship into a low-latency, telephony-class path and work alongside the speech-to-speech and LLM researchers. You report to the Head of AI in a deliberately flat team, with a possible technical lead path for the right person. What we’re looking for • Hands-on ownership of a neural audio codec or audio tokenizer, from training objective through to evaluation. • Experience training generative audio models with a clear personal contribution to the architecture and training, beyond data, evals or product glue. • Real industry experience taking a TTS model from research prototype to production scale, plus post-training work that made it reliable and stable. You can ship as well as publish. • Shipping into a low-latency telephony path, ASR experience alongside TTS, and first or second author papers in non-generic areas are a bonus. Salary and benefits Up to around $300K depending on experience, plus equity. Healthcare, including dental and vision. Logistics San Francisco office hybrid or fully remote. Start as soon as possible.