datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
TTS-Multilingual-Test-Set
Overview
To assess the multilingual zero-shot voice cloning capabilities of TTS models, we have constructed a test set encompassing 24 languages. This dataset provides both audio samples for voice cloning and corresponding test texts.
Specifically, the test set for each language includes:
100 distinct test sentences.
Audio samples from two speakers (one male and one female) carefully selected from the Mozilla Common Voice (MCV) dataset, intended for voice cloning.
Researchers can… See the full description on the dataset page: https://huggingface.co/datasets/MiniMaxAI/TTS-Multilingual-Test-Set.test-data-set-Arabic-letterasr-testset-kw-ja-v1
日本語 ASR アノテーション v1
重要：評価結果を報告する際の規約
本テストセットで評価結果を報告する際は、事前学習を含む学習にYouTubeの音声を使用したかどうかと、次の評価区分を必ず明記してください。
使用した場合：in-domain評価
使用していない場合：out-of-domain評価
この区分は、音声認識の評価結果を比較する上で重要です。最終的な追加学習だけでなく、使用するモデルの事前学習も含めて判断してください。
元データセットの音声に、人手で区間ごとの転記・タグ・採否を付けたデータです。音声の内容、ディレクトリ構成、ファイル名は元のままです。
ファイルと表示
**annotations.jsonl**：提出済みの全結果。1行が1音声です。
**metadata.jsonl**：HF表示用に自動生成したデータ。音声全体が不使用の行を除き、audio を file_name に置き換えています。
**data/**：採用した音声ファイル。
HFのビューアーは… See the full description on the dataset page: https://huggingface.co/datasets/bandad/asr-testset-kw-ja-v1.MMEdit-TestSet
MMEdit Test Set
A paired audio editing test set for text-guided audio manipulation evaluation, released with MMEdit.
Overview
This dataset contains 3,317 aligned triplets:
Component
Description
raw/
Source audio before editing
target/
Target audio after editing
content.jsonl
Editing instruction (caption) keyed by audio_id
Each sample is linked by a shared audio_id. For example, sample add_017221 corresponds to:
raw/add_017221.wav — original… See the full description on the dataset page: https://huggingface.co/datasets/CocoBro/MMEdit-TestSet.MCABSA_testsetContinuo-Testset
Continuo-Testset
Continuo-Testset is a benchmark for long-form, multi-speaker zero-shot speech generation. Each case asks a system to synthesize a complete speaker-attributed script as one recording, given a reference prompt for every speaker. The output is scored against a human-verified target recording.
This dataset accompanies an anonymous ICLR 2027 submission.
Statistics
Chinese (zh)
English (en)
Total
Cases
59
59
118
Single-speaker long-form
18… See the full description on the dataset page: https://huggingface.co/datasets/AnonyData/Continuo-Testset.asr-testset-kw-jaNVVSpeech-Challenge-Track1-Test-Set
NVVSpeech Challenge Track 1 Test Set
Track 1 test set for the NVVSpeech Challenge at ISCSLP 2026.
Task
Given a speech recording, produce a transcript that contains the spoken content and the non-verbal vocalization (NVV) tags at their corresponding positions.
Dataset Summary
Language
Samples
Chinese
985
English
961
Total
1,946
Files
.
├── README.md
├── SUBMISSION_GUIDE.txt
├── test.jsonl
├── ground_truth.jsonl
├──… See the full description on the dataset page: https://huggingface.co/datasets/NVVSpeech-Challenge/NVVSpeech-Challenge-Track1-Test-Set.X-Voice-TestsetX-Voice Multilingual Test Set
High-Fidelity Test Set for Multilingual Text-to-Speech across 30 Languages
This test set is built as part of the research: X-Voice: One Speaker, 30+ Languages with Zero-Shot Voice Cloning, serving as the evaluation benchmark for our model.
Dataset Summary
30 languages
European: bg (Bulgarian), cs (Czech), da (Danish), de (German), el (Greek), en (English), es (Spanish), et (Estonian), fi (Finnish), fr (French), hr (Croatian), hu (Hungarian), it… See the full description on the dataset page: https://huggingface.co/datasets/XRXRX/X-Voice-Testset.TTS-Multilingual-Test-Set
Overview
To assess the multilingual zero-shot voice cloning capabilities of TTS models, we have constructed a test set encompassing 24 languages. This dataset provides both audio samples for voice cloning and corresponding test texts.
Specifically, the test set for each language includes:
100 distinct test sentences.
Audio samples from two speakers (one male and one female) carefully selected from the Mozilla Common Voice (MCV) dataset, intended for voice cloning.
Researchers can… See the full description on the dataset page: https://huggingface.co/datasets/YoonSeon/TTS-Multilingual-Test-Set.kinyarwanda_cleaned_testset_verified_20HRSsortformer-diarization-test-set
Sortformer Diarization Test Set
100 real speech samples extracted from LibriSpeech test-clean for speaker diarization testing and benchmarking with NVIDIA Sortformer 4spk-v2 ONNX models.
Usage with Sortformer ONNX
from huggingface_hub import snapshot_download
import soundfile as sf
# Download the test set
dataset_path = snapshot_download("DimQ1/sortformer-diarization-test-set")
# Load audio
audio, sr = sf.read(f"{dataset_path}/audio/ls_real_000.wav")… See the full description on the dataset page: https://huggingface.co/datasets/DimQ1/sortformer-diarization-test-set.NVVSpeech-Challenge-Track2-Test-Set
NVVSpeech Challenge Track 2 Test Set
Track 2 test set for the NVVSpeech Challenge at ISCSLP 2026.
Task
Given a transcript containing one or more non-verbal vocalization (NVV) tags, synthesize speech that naturally realizes the requested NVVs while preserving intelligibility, naturalness, and audio quality.
Dataset Summary
Language
Samples
Chinese
800
English
800
Total
1,600
Files
.
├── README.md
├──… See the full description on the dataset page: https://huggingface.co/datasets/NVVSpeech-Challenge/NVVSpeech-Challenge-Track2-Test-Set.BandFlex_Testsetcat > README.md << 'EOF'
BandFlex Testset
Test sets for BandFlex bandwidth extension, built from VCTK and Expresso.
Download
VCTK Testset (~2.4 GB): included in this GitHub repo (VCTK_Testset/)
Full dataset (VCTK + Expresso, ~22 GB): hosted on Hugging Face
👉 https://huggingface.co/datasets/UmaSubhashiniRavuri1/BandFlex_Testset
Download from Hugging Face:
pip install -U huggingface_hub
hf download UmaSubhashiniRavuri1/BandFlex_Testset --repo-type=dataset… See the full description on the dataset page: https://huggingface.co/datasets/UmaSubhashiniRavuri1/BandFlex_Testset.cv10-uk-testset-clean
The cleaned Common Voice 10 (test set) that has been checked by a human for Ukrainian 🇺🇦
Overview
This repository contains the archive of Common Voice 10 (test set) with checked Ukrainian transcriptions and audios.
All audios have been checked by a human to be sure that they are correct.
This archive is used to test all ASR models listed here: https://github.com/egorsmkv/speech-recognition-uk
Community
Discord: https://bit.ly/discord-uds
Speech… See the full description on the dataset page: https://huggingface.co/datasets/Yehor/cv10-uk-testset-clean.corpus-siarad-test-setTTS-Multilingual-Test-Set
Overview
To assess the multilingual zero-shot voice cloning capabilities of TTS models, we have constructed a test set encompassing 24 languages. This dataset provides both audio samples for voice cloning and corresponding test texts.
Specifically, the test set for each language includes:
100 distinct test sentences.
Audio samples from two speakers (one male and one female) carefully selected from the Mozilla Common Voice (MCV) dataset, intended for voice cloning.
Researchers can… See the full description on the dataset page: https://huggingface.co/datasets/jeshica/TTS-Multilingual-Test-Set.amharic_cleaned_testset_verifiedkinyarwanda_cleaned_testset_verified_200HRSMSA_test_setPre-processed MSA data based on https://huggingface.co/datasets/mozilla-foundation/common_voice_16_1.
vi-en-ast-testsetamharic_cleaned_testset_fleurs_currentSageLM_testset_audioeval_framework_testsetAudio-Understanding-Test-Set
Audio Understanding Test Set
A structured dataset for evaluating audio understanding capabilities of multimodal AI models. Contains 137 test prompts across 22 categories, paired with a 20-minute voice sample and 49 completed model outputs from Gemini 3.1 Flash Lite.
Overview
Property
Value
Total prompts
137
Implemented (with prompt text)
49
Suggested (description only)
88
Completed outputs
49
Categories
22
Model under test… See the full description on the dataset page: https://huggingface.co/datasets/danielrosehill/Audio-Understanding-Test-Set.en-vi-ast-testsetwuw_testset1
Yougen/wuw_testset1
Wake-Up-Word (WUW) speech dataset, packed as WebDataset tar shards.
The input is a Kaldi-style data directory
(wav.scp, text, utt2spk, utt2dur, segments), where multiple
utterances share a long recording via the segments file.
To avoid duplicating audio, each tar sample corresponds to one full
recording. The utterance-level metadata (id / start / end / text / spk / duration) is stored in a JSON list inside that sample.
Downstream consumers slice the decoded… See the full description on the dataset page: https://huggingface.co/datasets/heimayuan/wuw_testset1.kinyarwanda_cleaned_testset_verifiedtest_setkinyarwanda_cleaned_testset_verified_10HRS