Plain-text fallback (for robots and lynx users)
The interactive versions of this page are /LINUX/ (Slackware Linux 3.0 terminal) and /MSDOS/ (MS-DOS 6.00 PC). The same content, as static text:
Morrie Doulaty, Ph.D.
Staff Research Engineer @ Google DeepMind | Cambridge & London, UK
About
I am a Staff Research Engineer at Google DeepMind in London, working on Speech LLMs, fully duplex voice agents, and agentic audio systems (modeling, evaluation, and orchestration). Previously I spent six years at Meta (London) as a Tech Lead Staff AI/ML Software Engineer and Senior Research Data Scientist, leading dialogue-quality metrics for audio-in/audio-out Llama models, training infrastructure for egocentric multimodal conversational understanding, and large-scale neural information retrieval. Earlier I was a Senior Applied Scientist at Microsoft and a Research Software Development Engineer at Microsoft Research in Munich, a Data Scientist at QuantumBlack (AI by McKinsey) in London, and a Research Intern at Google in New York. I hold a Ph.D. in Computer Science (Speech Technology and Machine Learning) from the University of Sheffield's Speech and Hearing Research Group, supervised by Prof. Thomas Hain and advised by Prof. Roger K. Moore.
Experience
- 2026/05 – now Staff Research Engineer, Google DeepMind, London, UK — Speech LLMs, fully duplex voice agents, agentic audio systems.
- 2021/05 – 2026/05 Tech Lead Staff AI/ML Software Engineer, Meta (Facebook), London, UK — led dialogue-quality metrics for audio-in/audio-out Llama models; led training infrastructure for egocentric multimodal conversational understanding.
- 2020/03 – 2021/05 Senior Research Data Scientist, Meta (Facebook), London, UK — large-scale neural-embedding information retrieval for semantic matching and copy detection; AI against misinformation on Facebook, Instagram and WhatsApp.
- 2018/06 – 2020/03 Senior Applied Scientist, Microsoft, Munich, Germany — deep-learning speech and language models for Cognitive Services.
- 2017/12 – 2018/06 Research Software Development Engineer, Microsoft Research, Munich, Germany — conversational AI models with Microsoft Office teams.
- 2017/02 – 2017/12 Data Scientist, QuantumBlack, AI by McKinsey & Company, London, UK — deep time-series models for predictive maintenance.
- 2017 – 2020 Visiting Researcher, Speech and Hearing Group, University of Sheffield, UK.
- 2015/07 – 2015/10 Research Intern, Google, New York, US — multi-style training of deep neural networks for ASR.
- 2013/02 – 2017/01 Natural Speech Technology Researcher, University of Sheffield, UK — multi-domain speech recognition; scientific computing cluster.
Education
- 2013 – 2017 Ph.D. in Computer Science (Speech Technology and Machine Learning), University of Sheffield, UK. Thesis: Methods for Addressing Data Diversity in Automatic Speech Recognition. Supervisor Prof. Thomas Hain, advisor Prof. Roger K. Moore.
- 2009 – 2012 M.Sc. in Computer Science (Intelligent Systems), University of Tabriz, Iran. Thesis: Document Classification Using Kernel Approaches.
- 2005 – 2009 B.Sc. (Hons) in Computer Science, University of Tabriz, Iran.
Selected publications
- Seide, F., Doulaty, M., et al. (2024). Speech ReaLLM – Real-time Streaming Speech Recognition with Multimodal LLMs by Teaching the Flow of Time. Interspeech 2024.
- Liu, X., …, Doulaty, M., et al. (2023). SynthVSR: Scaling up Visual Speech Recognition with Synthetic Supervision. CVPR 2023.
- Grauman, K., …, Doulaty, M., et al. (2022). Ego4D: Around the World in 3,000 Hours of Egocentric Video. Proc. CVPR 2022 (EgoVis Distinguished Paper Award).
- Doulaty, M., Hain, T. (2019). Latent Dirichlet Allocation Based Acoustic Data Selection for Automatic Speech Recognition. Proc. Interspeech 2019.
- Deena, S., Hassan, M., Doulaty, M., Saz, O., Hain, T. (2018). Recurrent Neural Network Language Model Adaptation for Multi-Genre Broadcast Speech Recognition and Alignment. IEEE Transactions on Audio, Speech and Language Processing.
- Doulaty, M., Rose, R., Siohan, O. (2016). Automatic Optimisation of Data Perturbation Distributions for Multi-Style Training in Speech Recognition. Proc. SLT 2016.
- Doulaty, M., Saz, O., Hain, T. (2015). Data-selective Transfer Learning for Multi-Domain Speech Recognition. Proc. Interspeech 2015.
- Doulaty, M. (2017). Methods for Addressing Data Diversity in Automatic Speech Recognition. Ph.D. thesis, The University of Sheffield.
Full list (18 papers, 2015–2024): cat cv/publications.txt in /LINUX/ or TYPE CV\PUBLICAT.TXT in /MSDOS/.
Awards
- 2024 Recognised as an Exceptional Talent under the UK Global Talent programme, endorsed by TechNation.
- 2022 Egocentric Vision (EgoVis) 2022/2023 Distinguished Paper Award for Ego4D (CVPR 2022).
- 2013 EPSRC-funded 3.5-year Ph.D. scholarship, Natural Speech Technology project (EP/I031022/1).
Contact
- Email: [email protected]
- Location: Cambridge / London, United Kingdom
- Curriculum vitae: mdoulaty_cv.pdf