大模型语音算法专家【外资 Remote 预算 高 高】120k以上
北京硕士及以上经验不限大模型TTS
RESPONSIBILITIES
• Research & Experimentation – Explore and develop new techniques across LLMs and audio models to improve reasoning, latency, and conversational quality in realtime systems.
• Model Training – Rapidly build and iterate on models and pipelines, turning research ideas into working prototypes. Innovate on paradigms, training methods, and inference.
• Evaluation & Benchmarking – Design novel evaluation frameworks, datasets, and metrics to measure performance on complex, real-world voice tasks.
• Bridge Research to Production – Collaborate closely with engineering to translate research insights into deployable systems.
• Human Feedback Loops – Develop methods to incorporate human evaluation into model improvement, especially for subjective conversational quality.
• Advance the Frontier – Stay at the cutting edge of ML research and bring new ideas into Retell’s product and infrastructure. REQUIRED
• Strong ML Research Background – You've worked on advanced ML problems (like LLM pre-training and post-training, transcription model training, TTS, or multimodal systems), either in industry or academia.
• Deep Technical Foundation – Comfortable with PyTorch, model architectures, and the math behind modern machine learning.
• Top Academic Background – Master's degree in CS, ML, AI or related field required; PhD preferred. Equivalent research-level engineering experience also considered.
YOU MIGHT THRIVE IF YOU
• Published or Awarded – First/co-author publications at top-tier venues (NeurIPS, ICML, ICLR, ACL, Interspeech, etc.) or notable competition awards are a strong plus.
• Experimental Mindset – You enjoy exploring open-ended problems and iterating quickly on ideas.
• Bridge Theory & Practice – You can translate research into systems that work in realworld environments.
• Startup-Ready – You thrive in fast-paced environments with high ownership and ambiguity.
• Collaborative & Clear Communicator – You can explain complex ideas and work cross-functionally to drive impact.