Georgi Tinchev

Georgi Tinchev

Scientist @ Amazon Alexa

University of Oxford

Biography

Hi! I am currently a Scientist at Amazon Research, working on generative modelling and speech disentanglement of text-to-speech systems on Alexa.


Previously I was a DPhil (PhD) student with Dr. Maurice Fallon at the University of Oxford as part of the Dynamics Robotics Group at the Oxford Robotics Institute. My research interests include SLAM, place recognition, keypoint detection and correspondence estimation, and generative deep learning, specifically diffusion models and transformers. I also interned as computer vision scientist at XYZ Reality working on correspondence estimation algorithms. Prior to that I interned as an applied scientist at Amazon Research working on speech synthesis for Alexa. I completed a MSc Artificial Intelligence degree at the University of Edinburgh and I hold a first class honours degree in computing science from the University of Aberdeen. My free time I spend playing volleyball, travelling and skiing.


Download my resumé.

Interests
  • SLAM/SfM
  • Keypoint detection in 2D/3D
  • Generative learning
  • Representation learning
  • Geometry
Education
  • PhD in Robotics and Computer Vision, 2021

    University of Oxford

  • MSc in Artificial Intelligence, 2016

    University of Edinburgh

  • BSc in Computer Science, 2015

    University of Aberdeen

Skills

Computer Vision

85%

Machine Learning

85%

Robotics

100%

Deep Learning

92%

Python

95%

Volleyball

84%

Experience

 
 
 
 
 
Amazon
Applied Scientist II
Amazon
Oct 2021 – Present London, UK

Responsibilities include:

  • Tech lead of a team on disentanglement technology for Alexa
  • Delivered machine learning models at scale to production
  • Reserach on state-of-the-art generative models
 
 
 
 
 
University of Oxford
PhD/DPhil student
University of Oxford
Apr 2016 – Apr 2021 Oxford, UK

Responsibilities include:

  • Developed SOTA algorithms for place recognition, odometry, navigation
  • Deployed a LIDAR loop closure detecting system
  • Supervised MSc students on 3D deep learning
 
 
 
 
 
XYZ Reality
Computer Vision Scientist
XYZ Reality
Oct 2020 – May 2021 London, UK

Responsibilities include:

  • Inventing and evaluating correspondence networks
  • Protyping proof-of-concept NeRF
  • Patenting SLAM technology
 
 
 
 
 
Amazon
Applied Scientist
Amazon
Nov 2019 – Aug 2020 Cambridge, UK

Responsibilities include:

  • Compared efficient generative models
  • Worked in a team to create a single universal vocoding model replacing 43 other models
  • Researched features for latent representaton of speech signals
 
 
 
 
 
Ikiji Ltd
Software Developer
Ikiji Ltd
Jun 2013 – Nov 2018 Aberdeen, UK
Designed solutions with web frameworks for multiple clients

Recent Publications

(2023). InstaLoc: One-shot Global Lidar Localisation in Indoor Environments through Instance Learning. In RSS.

PDF Cite DOI

(2023). Diffusion-based accent modelling in speech synthesis. In Interspeech.

PDF Cite Source Document

(2023). Modelling low-resource accents without accent-specific TTS frontend. In ICASSP.

PDF Cite Source Document DOI

(2021). Universal Neural Vocoding with Parallel Wavenet. In ICASSP.

PDF Cite DOI

(2020). 𝕏 Resolution Correspondence Networks. In BMVC.

PDF Cite Code

(2020). SKD: Keypoint Detection for Point Clouds using Saliency Estimation. In RAL/ICRA.

PDF Cite Slides Video DOI

(2020). Online LiDAR-SLAM for Legged Robots with Robust Registration and Deep-Learned Loop Closure. In ICRA.

PDF Cite Video DOI

(2018). Predicting Alignment Risk to Prevent Localization Failure. In ICRA.

PDF Cite Video DOI

Contact