AIML - Machine Learning Research (Speech-to-Speech Translation) - TMI

The Machine Translation and Multilingual Intelligence team is looking for outstanding engineers and scientists to build the next generation of speech-to-speech translation technology, so our users can talk naturally with anyone, in any language, in real time.

You will take part in the next revolution at the confluence of machine translation, speech technology and human-computer interaction. We are building end-to-end speech-to-speech translation: models that listen, translate and speak in a single system, with low latency and natural, expressive output.

You will:

• Have a significant impact by shipping speech-to-speech translation technology to millions of real-world users.
• Enable users to communicate seamlessly across languages, in conversation, on calls and through their devices.
• Address the practical problems of real-time translation at scale: translation quality, latency and streaming, natural and expressive speech output, robustness to real-world audio, and evaluation that reflects the user experience.

We are looking for people with a deep, real passion for technology who enjoy collaborating with researchers, engineers and designers to put this technology in the hands of millions of customers.

Conduct independent applied research and development in speech-to-speech translation, from modelling through to evaluation and shipping. Design, train and improve large-scale multimodal models that combine speech understanding, translation and speech generation. Advance streaming and simultaneous translation, balancing translation quality against latency. Improve the quality, naturalness and expressiveness of translated speech. Define and run automatic and human evaluations for speech translation quality, latency and speech quality.

M.S. or PhD in Computer Science or a related field. Experience with machine learning for speech recognition, machine translation, NLP or speech synthesis. Hands-on experience with deep learning toolkits such as PyTorch, JAX or TensorFlow.

Research or product experience in speech-to-speech or speech-to-text translation, especially end-to-end, streaming or simultaneous approaches. Experience in text-to-speech and speech generation, such as neural TTS, speech tokenizers and audio codecs, expressive or multilingual synthesis. Experience with large language models or multimodal (audio + text) foundation models, including pre-training, fine-tuning or reinforcement learning. Experience designing and deploying real-world, large-scale, user-facing speech recognition, translation or synthesis systems, including on-device or low-latency inference. Familiarity with speech and translation evaluation, such as WER, BLEU/chrF/COMET-style metrics, MOS and side-by-side human evaluation, and latency metrics. Ability to formulate a research problem, then design, run experiments and implement solutions in Python and C/C++. Publications in relevant venues (e.g. Interspeech, ICASSP, ACL, EMNLP, NeurIPS, ICML) are a plus. Excellent spoken and written communication skills.