Research Engineer/Research Scientist, Audio
San Francisco, CAFull-TimeMid-levelResearch
About Anthropic
- Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.
- Anthropic’s Audio team pushes the boundaries of what's possible with audio with large language models. We care about making safe, steerable, reliable systems that can understand and generate speech and audio, prioritizing not only naturalness but also steerability and robustness. As a researcher on the Audio team, you'll work across the full stack of audio ML, developing audio codecs and representations, sourcing and synthesizing high quality audio data, training large-scale speech language models and large audio diffusion models, and developing novel architectures for incorporating continuous signals into LLMs.
- Our team focuses primarily but not exclusively on speech, building advanced steerable systems spanning end-to-end conversational systems, speech and audio understanding models, and speech synthesis capabilities. The team works closely with many collaborators across pretraining, finetuning, reinforcement learning, production inference, and product to get advanced audio technologies from early research to high impact real-world deployments.
You may be a good fit if you
- Have hands-on experience with training audio models, whether that's conversational speech-to-speech, speech translation, speech recognition, text-to-speech, diarization, codecs, or generative audio models
- Genuinely enjoy both research and engineering work, and you'd describe your ideal split as roughly 50/50 rather than heavily weighted toward one or the other
- Are comfortable working across abstraction levels, from signal processing fundamentals to large-scale model training and inference optimization
- Have deep expertise with JAX, PyTorch, or large-scale distributed training, and can debug performance issues across the full stack
- Thrive in fast-moving environments where the most important problem might shift as we learn more about what works
- Communicate clearly and collaborate effectively; audio touches many parts of our systems, so you'll work closely with teams across the company
- Are passionate about building conversational AI that feels natural, steerable, and safe
- Care about the societal impacts of voice AI and want to help shape how these systems are developed responsibly
Strong candidates may also have experience with
- Large language model pretraining and finetuning
- Training diffusion models for image and audio generation
- Reinforcement learning for large language models and diffusion models
- End-to-end system optimization, from performance benchmarking to kernel optimization
- GPUs, Kubernetes, PyTorch, or distributed training infrastructure
Representative projects
- Training state-of-the art neural audio codecs for 48 kHz stereo audio
- Developing novel algorithms for diffusion pretraining and reinforcement learning
- Scaling audio datasets to millions of hours of high quality audio
- Creating robust evaluation methodologies for hard-to-measure qualities such as naturalness or expressiveness
- Studying training dynamics of mixed audio-text language models
- Optimizing latency and inference throughput for deployed streaming audio systems
- The annual compensation range for this role is listed below.
- For sales roles, the range provided is the role’s On Target Earnings ("OTE") range, meaning that the range includes both the sales commissions/sales bonuses target and annual base salary for the role.
How we're different
- We believe that the highest-impact AI research will be big science. At Anthropic we work as a single cohesive team on just a few large-scale research efforts. And we value impact — advancing our long-term goals of steerable, trustworthy AI — rather than work on smaller and more specific puzzles. We view AI research as an empirical science, which has as much in common with physics and biology as with traditional efforts in computer science. We're an extremely collaborative group, and we host frequent research discussions to ensure that we are pursuing the highest-impact work at any given time. As such, we greatly value communication skills.
- The easiest way to understand our research directions is to read our recent research. This research continues many of the directions our team worked on prior to Anthropic, including: GPT-3, Circuit-Based Interpretability, Multimodal Neurons, Scaling Laws, AI & Compute, Concrete Problems in AI Safety, and Learning from Human Preferences.
