Atli Thor Sigurgeirsson

PhD Student @ The University of Edinburgh | Natural language processing and Text-To-Speech

About

Atli Thor Sigurgeirsson

Hey! I work on speech generation models! I’m interested in controllability and how we interact with complex generative models.

I’m currently wrapping up my PhD at CSTR, University of Edinburgh. I'm actively looking for opportunities starting August 2025!

HuggingFace LinkedIn Google Scholar

Research

RepeaTTS: Towards feature discovery through repeated fine-tuning

A Sigurgeirsson, S King, SSW 2025

Controllable speaking styles using a large language model

A Sigurgeirsson, S King, ICASSP 2024

Do Prosody Transfer Models Transfer Prosody?

A Sigurgeirsson, S King, ICASSP 2023

Just Because We Camp, Doesn't Mean We Should: The Ethics of Modelling Queer Voices

A Sigurgeirsson, E Ungless, Interspeech 2024

A Human-in-the-Loop Approach to Improving Cross-Text Prosody Transfer

H Maurya, A Sigurgeirsson, Interspeech 2024

Corpora

Talrómur: Large general use TTS corpus

Atli Þór Sigurgeirsson, Þorsteinn Daði Gunnarsson, Gunnar Thor Örnólfsson, Ragnheiður Þórhallsdóttir, Eydís Huld Magnúsdóttir, Jón Guðnason, 2021

Talrómur 3: Multi-speaker emotive TTS corpus

GT Örnólfsson, AÞ Sigurgeirsson, AB Nikulásdóttir, D Schnell, 2024

Available TTS models

Emotive Icelandic

An Icelandic TTS model that can synthesise speech according to natural language descriptions of the output.

RepeaTTS

The model proposed in my paper: RepeaTTS.

Contact

Email: atlisigurgeirsson@gmail.com