About the role#
Centific AI Research is seeking a PhD Research Intern to design and evaluate speech-first models. You will focus on Spoken Language Models (SLMs) that reason over audio and interact conversationally, moving ideas from initial prototypes to practical demonstrations. You will work alongside scientists and engineers to deliver measurable impact through end-to-end speech dialogue systems and speech-aware LLMs.
What you'll do#
- Prototype a conversational SLM using an SSL speech encoder and a compact adapter on an existing LLM.
- Create a data recipe blending conversational speech with instruction-following corpora, including targeted ablations.
- Build an evaluation harness for ASR, ST, and SLU tasks, covering speech QA and streaming metrics like latency and stability.
- Design and evaluate speech-first models, focusing on modality alignment, efficient speech tokenization, and temporal compression.
- Document research findings, including goals, design choices, and steps for productionization.
What you'll need#
- Current PhD candidate in Computer Science, Electrical Engineering, or a related field with research in speech, audio ML, or multimodal LMs.
- Fluency in Python and PyTorch, with hands-on experience in GPU training and familiarity with torchaudio or librosa.
- Working knowledge of modern sequence models, such as Transformers or SSMs, and training best practices.
- Demonstrated depth in at least one of the following: discrete speech tokens and temporal compression, modality alignment to LLMs via adapters, or post-training and instruction tuning for speech tasks.
- Strong experimentation habits, including clean code, reproducibility, and clear reporting.
Location & details#
- Location: Remote (United States), with a preference for candidates in Washington.
- Type: Full-time (40 hours per week).
- Term: Rolling.
- Compensation: $45 per hour.
