Research Scientist, Gemini Audio i18n, DeepMind

• Create comprehensive evaluation sets and benchmarks to measure audio-to-audio (A2A) model performance across international languages, accents, and regional dialects.
• Propose, prototype, and evaluate novel modeling techniques to improve A2A audio understanding, dialog, and audio generation capabilities with a focus on scalability.
• Identify performance gaps in current multilingual audio models and collaborate with cross-functional research and engineering teams to deploy solutions.

Minimum qualifications:

• Bachelor's degree in Computer Science, Speech Recognition, Computational Linguistics, a related technical field, or equivalent practical experience.
• Experience conducting research or development in Speech Recognition, Text-to-Speech (TTS), or Large Language Models (LLMs).
• Experience coding in Python or C++ and using deep learning frameworks such as PyTorch, JAX, or TensorFlow.
• Experience working with audio data, speech processing, or multilingual datasets.

Preferred qualifications:

• Master's degree or Ph.D. in Automatic Speech Recognition (ASR), Text-to-Speech (TTS), Computational Linguistics, or a related field.
• 3 years of experience in Large Language Models (LLMs) or multimodal foundation models.
• Experience developing audio-to-audio (A2A) architectures, end-to-end speech models, or spoken dialog systems.
• Experience scaling speech models across international languages, accents, or low-resource locales.
• Publication record in speech or machine learning conferences (e.g., ICASSP, INTERSPEECH, NeurIPS, or ACL).

• Bachelor's degree in Computer Science, Speech Recognition, Computational Linguistics, a related technical field, or equivalent practical experience.
• Experience conducting research or development in Speech Recognition, Text-to-Speech (TTS), or Large Language Models (LLMs).
• Experience coding in Python or C++ and using deep learning frameworks such as PyTorch, JAX, or TensorFlow.
• Experience working with audio data, speech processing, or multilingual datasets.