About Me
I like to solve real-world problems, and most of the time, I use neural networks. Particularly, I am fascinated by speech and designing systems that aim to make sense of it the same as humans effortlessly do all the time.
This interest emerged while doing research in the Neurotherapeutics lab at the Institute of Cognitive Neuroscience, where I witnessed what aphasia is for the first time. Fundamentally, making sense of the unpredictable speech patterns characteristic of aphasic speakers requires a deeper understanding of speech, factoring in the ability of individuals to adapt to previously unheard speech patterns.
'Adaptability to something new' is one of the missing abilities of current AI systems. My lessons learned in trying to make sense of aphasic speech forced me to think in new directions. They helped me question commonly based assumptions in tech about speech. These assumptions find their way into the implementation of current speech systems in a field that seems overly dominated by engineers and flashy benchmarks to no avail for those who need adaptive systems the most.
You could be forgiven for thinking this is quite a niche research area to focus on. After all, datasets of aphasic speech are scarce, exist for a few languages and might have restricted access only to clinical institutions. Also, there are not many publications on speech recognition in aphasia compared to the vast field of ASR. However, conceiving a system that can make sense of aphasic speech not only forces us to re-think current strategies in signal processing, tokenisation or training paradigms in deep learning but also forces us to step deep into the realm of intelligence where humans, especially clinicians and speech and language therapists, champion the art of making sense of speech even if it appears to be broken.
Hence, shifting the current conversational AI efforts to aiming at this more complex problem has a better chance of yielding a technological breakthrough. It might look like a niche field, yet there is more gain from solving it, not just for people with aphasia but for the rest of us. Perhaps it is time to use this fact and turn the tables, pouring considerably more funding towards research in aphasia and attracting top talent in AI.
So here I am, trying to turn the tables to making sense of speech.
Research Interests
Areas I explored, I am exploring and willing to explore:
- Deep Learning
- Reinforcement Learning
- Active Inference
- Generative Models
- TinyML
Background
Admittedly, I have a very unorthodox background:
- BSc Mathematics & Economics, LSE
- MSc Computer Games & Entertainment, Goldsmiths
- PhD Institute of Cognitive Neurosience, UCL