Machine Learning Engineer, Speech/Audio

attention labs; San Francisco, Northern (CA, KY); Hybrid; USD 150,000 - 200,000; Full time

Job summary

attention labs is building the audio modeling stack in a fast-moving startup setting in San Francisco. You will own end-to-end lifecycle from data to production inference, shipping models that recognize who is being spoken to in real time.

You will turn research into dependable inference under real-world noise and latency constraints, design evaluation harnesses, and collaborate with founders to shape next steps for the product's audio intelligence.

Qualifications

• Strong applied ML experience with audio, speech, or signal-heavy models.
• Ability to own a model from data collection to production inference, not just notebooks.
• Fluency with modern deep learning tooling such as PyTorch and rigorous measurement.

Responsibilities

• Own the audio model lifecycle end to end: data, features, architecture, training, evaluation, and production inference.
• Build and improve models behind addressee detection and real-time decisions on who is being spoken to.
• Turn research prototypes into fast, dependable inference under real-world noise and accents.
• Design evaluation harnesses and datasets to honestly measure impact of changes.
• Work directly with founders and early customers to shape next model objectives.

Skills

Audio/speech ML

Production model lifecycle

PyTorch

Latency awareness

Tools

PyTorch

Job description

San Francisco · Full-time · Machine Learning

You will own the audio modeling stack end to end, from raw signal to production inference. You are the person who ships the models that decide, in real time, who a voice agent is being spoken to.

• Own the audio model lifecycle end to end: data, features, architecture, training, evaluation, and the inference path that runs in production.
• Build and improve the models behind addressee detection, deciding whether speech is meant for the agent or for someone else in the room.
• Turn research prototypes into fast, dependable inference that holds up under real world noise, overlap, and accents.
• Design the evaluation harness and datasets that tell us, honestly, whether a change made the product better.
• Work directly with founders and early customers to shape what the model needs to do next.

What we are looking for

• Strong applied machine learning experience with audio, speech, or other signal-heavy models.
• Comfort owning a model from data collection through to something running in production, not just a notebook.
• Fluency with modern deep learning tooling such as PyTorch, and the discipline to measure before and after every change.
• A bias toward shipping, and toward the smallest experiment that answers the question.
• Enough systems sense to care about latency, memory, and how a model behaves under load.
• Bonus: experience with real time audio, source separation, or on-device inference.

About attention labs

attention labs is early: a small team defining a new category at the intersection of speech, cognitive neuroscience, and machine learning. We work from San Francisco, Toronto, and Memphis, and we are remote-friendly for the right person.