The Blizzard Challenge 2025

The nineteenth Blizzard Challenge focused on speech synthesis for Bildts, a low resource language variety in the Netherands with approximately 6,000 native slpeakers. The main task provided a single-speaker 7-hour Bildts dataset and required<br/>participants to build a voice from this dataset, restricted to using only publicly available data and pre-trained models. Listeners evaluated synthesised samples for Bildts-likeness, naturalness, appropriateness, overall signal quality, and word-level quality (using RPT-like feedback). The optional zero-shot task provided 10 seconds of reference speech from two speakers; participants built voices matching these speakers without constraints on data or models. Listeners evaluated the main task for Bildts-likeness, naturalness, appropriateness, overall signal quality, and word-level errors, while the zero-shot task focused on Bildts-likeness and speaker similarity. Seven teams participated in the main task and five in the zero-shot task.

Authors

Publication Details

Published
2025-08-27
DOI
https://doi.org/10.21437/blizzard.2025-1
Primary Topic
Speech Recognition and Synthesis
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

The Blizzard Challenge 2025

Jelske Dijkstra, Igor Marchenko, Matt Coler, Vass Verkhodanova et al.
Speech Recognition and Synthesis
article

The Blizzard Challenge 2025

Jelske Dijkstra, Igor Marchenko, Matt Coler, Vass Verkhodanova, Sébastien Le Maguer, Phat Do
article en

Abstract

The nineteenth Blizzard Challenge focused on speech synthesis for Bildts, a low resource language variety in the Netherands with approximately 6,000 native slpeakers. The main task provided a single-speaker 7-hour Bildts dataset and required<br/>participants to build a voice from this dataset, restricted to using only publicly available data and pre-trained models. Listeners evaluated synthesised samples for Bildts-likeness, naturalness, appropriateness, overall signal quality, and word-level quality (using RPT-like feedback). The optional zero-shot task provided 10 seconds of reference speech from two speakers; participants built voices matching these speakers without constraints on data or models. Listeners evaluated the main task for Bildts-likeness, naturalness, appropriateness, overall signal quality, and word-level errors, while the zero-shot task focused on Bildts-likeness and speaker similarity. Seven teams participated in the main task and five in the zero-shot task.

Openalex Percentile: Top 17%
Speech Recognition and Synthesis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.