</> Code with Papers by Syntology

Papers › VOXLINGUA107: A DATASET FOR SPOKEN LANGUAGE RECOGNITION

VOXLINGUA107: A DATASET FOR SPOKEN LANGUAGE RECOGNITION

25 Nov 2020archive 2025-07-28

Jorgen Valk, Tanel Alumae

This paper investigates the use of automatically collected web audio data for the task of spoken language recognition. We generate semi-random search phrases from language-specific Wikipedia data that are then used to retrieve videos from YouTube for 107 languages. Speech activity detection and speaker diarization are used to extract segments from the videos that contain speech. Post-filtering is used to remove segments from the database that are likely not in the given language, increasing the proportion of correctly labeled segments to 98%, based on crowd-sourced verification. The size of the resulting training set (VoxLingua107) is 6628 hours (62 hours per language on the average) and it is accompanied by an evaluation set of 1609 verified utterances. We use the data to build language recognition models for several spoken language identification tasks. Experiments show that using the automatically retrieved training data gives competitive results to using hand-labeled proprietary datasets. The dataset is publicly available.

PaperPDF

Code

No code repository is listed for this paper in the archive or in Syntology's graph.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Action DetectionActivity DetectionLanguage IdentificationSpeaker DiarizationSpoken language identificationspeaker-diarization

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Spoken language identification KALAKA-3 Model on the automatically filtered (cleaned) data EC 0.022 #1 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the automatically filtered (cleaned) data EO 0.058 #1 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the automatically filtered (cleaned) data PC 0.041 #1 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the automatically filtered (cleaned) data PO 0.056 #1 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the noisy data EC 0.033 #2 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the noisy data EO 0.059 #2 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the noisy data PC 0.055 #2 of 2 Archive leaderboard report
Spoken language identification KALAKA-3 Model on the noisy data PO 0.083 #2 of 2 Archive leaderboard report
Spoken language identification LRE07 CNN-LDE 10 sec 2.61 #1 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-LDE 3 sec 8.25 #1 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-LDE 30 sec 1.16 #1 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-LDE Average 4.00 #1 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-SAP 10 sec 2.49 #2 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-SAP 3 sec 8.59 #2 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-SAP 30 sec 1.09 #2 of 9 Archive leaderboard report
Spoken language identification LRE07 CNN-SAP Average 4.06 #2 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (cleaned data) 10 sec 3.14 #3 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (cleaned data) 3 sec 9.39 #3 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (cleaned data) 30 sec 1.90 #3 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (cleaned data) Average 4.81 #3 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (noisy data) 10 sec 3.33 #4 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (noisy data) 3 sec 10.58 #4 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (noisy data) 30 sec 1.72 #4 of 9 Archive leaderboard report
Spoken language identification LRE07 Resnet34 (noisy data) Average 5.21 #4 of 9 Archive leaderboard report
Spoken language identification LRE07 Fusion of models 10 sec 4.54 #5 of 9 Archive leaderboard report
Spoken language identification LRE07 Fusion of models 3 sec 15.29 #5 of 9 Archive leaderboard report
Spoken language identification LRE07 Fusion of models 30 sec 1.30 #5 of 9 Archive leaderboard report
Spoken language identification LRE07 Fusion of models Average 7.04 #5 of 9 Archive leaderboard report
Spoken language identification LRE07 GMM-MMI 10 sec 5.90 #6 of 9 Archive leaderboard report
Spoken language identification LRE07 GMM-MMI 3 sec 17.28 #6 of 9 Archive leaderboard report
Spoken language identification LRE07 GMM-MMI 30 sec 2.10 #6 of 9 Archive leaderboard report
Spoken language identification LRE07 GMM-MMI Average 8.42 #6 of 9 Archive leaderboard report
Spoken language identification LRE07 Phonotactic 10 sec 6.28 #7 of 9 Archive leaderboard report
Spoken language identification LRE07 Phonotactic 3 sec 18.59 #7 of 9 Archive leaderboard report
Spoken language identification LRE07 Phonotactic 30 sec 1.34 #7 of 9 Archive leaderboard report
Spoken language identification LRE07 Phonotactic Average 8.73 #7 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector DNN 10 sec 7.84 #8 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector DNN 3 sec 19.67 #8 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector DNN 30 sec 3.31 #8 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector DNN Average 10.27 #8 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector 10 sec 11.93 #9 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector 3 sec 26.04 #9 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector 30 sec 4.52 #9 of 9 Archive leaderboard report
Spoken language identification LRE07 Kaldi i-vector Average 14.17 #9 of 9 Archive leaderboard report
Spoken language identification VOXLINGUA107 Noisy 0..5sec 12.3 #1 of 2 Archive leaderboard report
Spoken language identification VOXLINGUA107 Noisy 5..20sec 6.1 #1 of 2 Archive leaderboard report
Spoken language identification VOXLINGUA107 Noisy Average 7.1 #1 of 2 Archive leaderboard report
Spoken language identification VOXLINGUA107 Cleaned 0..5sec 13.4 #2 of 2 Archive leaderboard report
Spoken language identification VOXLINGUA107 Cleaned 5..20sec 6.6 #2 of 2 Archive leaderboard report
Spoken language identification VOXLINGUA107 Cleaned Average 7.6 #2 of 2 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections