ML Engineer & Research Scientist

About the Role

We are hiring an ML Engineer & Research Scientist to own a critical end-to-end research loop in speech AI: defining what it means for machines to talk like humans, measuring current gaps, and closing them. This is a highly autonomous role where you will work directly with founders on open problems across recognition, synthesis, conversation, emotion, languages, and accents.

Key Responsibilities

• Identify Problems: Listen to model failures in real-world scenarios (noise, accents) and define clear problem statements where current models fall short.
• Define Behaviour: Precisely specify how a speech model should behave for each identified gap, establishing targets for labs worldwide.
• Design Evals: Create test sets, automatic metrics, and human evaluation protocols to measure progress toward 'human-like' voice. Ensure evals are rigorous against gaming (using held-out data) and cover diverse conditions like code-mixing and noise.
• Define Datasets: Specify recording requirements for operations teams covering languages and accents in the Global South, ensuring high-quality labeling.
• Train Models: Train and fine-tune open ASR, TTS, and speech-to-speech models on your defined data. Prove improvements through clean ablations to show gains come from new behavior/data, not test-set tuning.
• Publish & Open Source: Release results as papers (Interspeech, ICASSP) and open-source code with model cards so the community can build upon them.

Impact

Your work directly shapes how labs build speech models. Every eval you ship decides what gets measured; every model you train shows the field what is possible in voice AI.