Resources
Resources
VoiceMOS Challenge 2022: The VoiceMOS Challenge 2022
VoiceMOS Challenge 2023: The VoiceMOS Challenge 2023: Zero-shot Subjective Speech Quality Prediction for Multiple Domains
VoiceMOS Challenge 2024: The VoiceMOS Challenge 2024: Beyond Speech Quality Prediction
AudioMOS Challenge 2025: The AudioMOS Challenge 2025
VoiceMOS Challenge 2026: The VoiceMOS Challenge 2026: Evaluating Speech Enhancement, Emotional TTS and Accented TTS Systems
INTERSPEECH 2025 tutorial: Automatic Quality Assessment for Speech and Beyond by Wen-Chin Huang, Erica Cooper, Jiatong Shi [Slides]
The BC2023 dataset: Track 1 dataset for the VoiceMOS Challenge 2023. [Link]
The SVCC2023 dataset: Track 2 dataset for the VoiceMOS Challenge 2023. [Link]
The TMHINT-QI(S) dataset: Track 3 dataset for the VoiceMOS Challenge 2023. [Link]
The zoomed-in BVCC dataset: Track 1 dataset for the VoiceMOS Challenge 2024. [Link][Paper]
The SingMOS dataset: Track 2 dataset for the VoiceMOS Challenge 2024. [Link][Paper]
Track 3 dataset for the VoiceMOS Challenge 2024. [Link (training set)][Link (test set)]
The MusicEval dataset: Track 1 dataset for the AudioMOS Challenge 2025. [Link][Paper]
Track 2 dataset for the AudioMOS Challenge 2025. [Link (test set audio)][Link (test set audio)]
Track 3 dataset for the AudioMOS Challenge 2025.
Training & development sets: [Link] [Labels (development set)]
The URGENT challenge 2026 dataset: Track 1 dataset for the VoiceMOS Challenge 2026. [Paper]
Track 2 dataset for the VoiceMOS Challenge 2026.
Note: We are working on making the dataset public. Before then, to obtain the dataset, one can follow the process below. However, we will only be doing this until Sep 30th 2026.
To obtain the data, please download the license file here: https://drive.google.com/file/d/1XYFRWCzKHRelz6KGZJlTsR-PAdacJ__d/view?usp=share_link.
Then, fill it out and email it to ecooper@nict.go.jp. The data links will be shared once the information in your license has been checked and confirmed.
CodecMOS-Accent: Track 3 dataset for the VoiceMOS Challenge 2026. [Paper]
VCTK distribution, containing natural samples from VCTK.
SYN distribution, containing the synthesized samples.
Note: We are working on making the dataset public. Before then, to obtain the dataset, one can follow the process below. However, we will only be doing this until Sep 30th 2026.
The To obtain the data, please download the license file here: https://drive.google.com/file/d/1wXHW784k5KQwik0YacXDjG95vo3zm5Ak/view?usp=sharing
Then, fill it out and email it to ecooper@nict.go.jp. The data links will be shared once the information in your license has been checked and confirmed.
SSL-MOS: Baseline system for the VoiceMOS Challenge 2022 and 2023. [Github link][Paper]
LDNet: Baseline system for the VoiceMOS Challenge 2022. [Github link][Paper]
MOSANet: Baseline system for the VoiceMOS Challenge 2022. [Github link][Paper]
UTMOS: Winning system of the VoiceMOS Challenge 2022, and the baseline system for the VoiceMOS Challenge 2023. [Github link][Paper]
RAMP: Top system of the VoiceMOS Challenge 2023 and the VoiceMOS Challenge 2024. [Github link][Paper]
VoiceMOS Challenge 2026 baseline systems [Github link]