urgent-challenge/urgent2024_official

Dataset Description

This dataset contains the official validation, non-blind test and blind test data used during the 2024 URGENT Speech Enhancement Challenge (https://urgent-challenge.github.io/urgent2024/), an official NeurIPS 2024 Competition. The dataset is designed to evaluate the performance of speech enhancement systems in handling various distortions and input conditions (e.g., varying sampling frequencies and encoding formats such as MP3/FLAC).

The dataset is divided into three splits:

• Validation Set: This split is used in the validation phase of the challenge for tuning the models and hyperparameters. It contains 1000 simulated samples (~1.5 hours) with paired clean and noisy speech, along with their meta information such as SNR levels, distortion configurations, transcript, and so on.
• Non-blind Test Set: This split is used in the non-blind phase of the challenge for evaluating the performance of the models. It contains 1000 simulated samples (~2 hours) with paired clean and noisy speech, along with their meta information such as SNR levels, distortion configurations, transcript, and so on.
• Blind Test Set: This split is used for the final blind test phase. It contains 500 samples of simulated samples and 500 real recordings (~2.5 hours in total). In contrast to other splits, no labels or meta information are provided for the blind test set.

All speech samples in this dataset are in English, with a single microphone channel. The dataset is intended to be used for evaluating speech enhancement systems, with a focus on robustness and generalizability across different conditions.

Example Usage

Each split of the dataset can be loaded individually using the datasets library. Below is an example of how to load the validation set:

from datasets import load_dataset

data = load_dataset("urgent-challenge/urgent2024_official", "validation")

# Load a single sample
sample = data["validation"][100]
print(sample)

# Iterate over all samples
# for idx, sample in enumerate(data["validation"]):
#     print(sample)

This will generate the following output:

{
  'id': 'fileid_14512',
  'sampling_rate': 32000,
  'noisy_audio': {
    'path': None,
    'array': array([0.01727295, 0.01797485, 0.01831055, ..., 0.0032959 , 0.00238037, 0.01062012]),
    'sampling_rate': 32000
  },
 'clean_audio': {
    'path': None,
    'array': array([ 0.00000000e+00,  0.00000000e+00,  0.00000000e+00, ..., -6.10351562e-05, -6.10351562e-05, -6.10351562e-05]),
    'sampling_rate': 32000
  },
 'transcript': 'He burnt out his tires after the drag race.',
 'snr_dB': 8.524861335754395,
}

Similarly, you can load the non-blind and blind test sets by changing the split name in the load_dataset function.

from datasets import load_dataset

data = load_dataset("urgent-challenge/urgent2024_official", "test_nonblind")

# Load a single sample
sample = data["validation"][100]
print(sample)

data = load_dataset("urgent-challenge/urgent2024_official", "test_blind")

# Load a single sample
sample = data["test"][100]
print(sample)

Note that the data structure in the blind test split slightly differs from the other splits, as some of the fields are not available:

{
  'id': 'fileid_101',
  'sampling_rate': 24000,
  'noisy_audio': {'path': None, 'array': array([-0.01263428, -0.01339722, -0.00991821, ..., -0.01495361, -0.00637817, -0.00637817]), 'sampling_rate': 32000},
}

More information and analysis

For more information about the dataset, including the data generation process, evaluation metrics, and analysis of the results, please refer to the official challenge website and our analysis paper (to be released).

Acknowledgment and license information

The dataset was created based on the following source speech, noise, and room impulse response (RIR) data, which are publicly available with varying licenses:

Source speech

Source	License
LibriVox in DNS5 Challenge	CC BY 4.0
LibriTTS	CC BY 4.0
CommonVoice 11.0 (English portion)	CC0
VCTK	ODC-BY
SLR83	CC BY-SA 4.0
CHiME-6	CC BY-SA 4.0
VOiCES	CC BY 4.0
EasyCom	CC BY-NC 4.0
Audios extracted from Youtube videos	CC BY 4.0

Expand to see a full list of Youtube audio data used in this competition

2ziV6vH-5wo, 4SzRZvEvyr8, 7XcyE1Rrp8g, 5ZqqXHz_xyo, 617jBfdO8MQ, AOvV9m7feKo, JDAYv4tv7yk, JQkewbO-qf4, rj27kDNtl2c, uB_ykR_yjnY, 4G1__QdEp0U, AbixsuKqfLQ, G3zK1aB_zU4, LWC3dS8EEDM, LZnNzQse2OA, QCltA8fDdQw, RaQ96aK4f58, WFAf2kEEBaI, c6UOJfoenss, kBJ2yTaq-UU, klvdARnvpjg, kmQBICeWhAQ, mWF5GXvfP-4, q_Tjw0SnRu8, ud0Aqc3lqLU, wr1hs6geXyk, omhUqE0SANk, d5NEhm7ZOyc, xw39PKAXkdE, qQrY8ZjSCnQ, GLeoET3gZWg, i_r3gCZrdSc, omhUqE0SANk, qQrY8ZjSCnQ, COls47WPPNk, qQrY8ZjSCnQ, xw39PKAXkdE, In35ZJEGtu8, bLVLhLMoV70, pkc-AJTfNDY, nIWnC9RMW08, GLeoET3gZWg, qQrY8ZjSCnQ, FtN8XzP8ZWg, 9up7LiqU9u4, i_r3gCZrdSc, omhUqE0SANk, nIWnC9RMW08, zWDJojt_KTU, AAckOJEBODk, VE0v1gsOMlw, kDDEQUfMT_8, d5NEhm7ZOyc, DUAdarwNCRs, qQrY8ZjSCnQ, 9jiyee7zCXo, 9up7LiqU9u4, pauj3PKE6L4, GLeoET3gZWg, 5YrhJMiyaPM, qQrY8ZjSCnQ, 5YrhJMiyaPM, POXAC1WpDBU, FtN8XzP8ZWg, qQrY8ZjSCnQ, DUAdarwNCRs, pauj3PKE6L4, qQrY8ZjSCnQ, 4geYwpsHQd8, 2r2BRN2tAFc, pauj3PKE6L4, xw39PKAXkdE, ZtUv67HHw6g, POXAC1WpDBU, Ab6ryHD_ahQ, zWDJojt_KTU, nIWnC9RMW08, d5NEhm7ZOyc, COls47WPPNk, 9up7LiqU9u4, omhUqE0SANk, 9R4O3bzONQQ, l6L2xaYRm7U, nIWnC9RMW08, qQrY8ZjSCnQ, zWDJojt_KTU, COls47WPPNk, GLeoET3gZWg, nIWnC9RMW08, mL6-d5mTQuM, qQrY8ZjSCnQ, bLVLhLMoV70, qQrY8ZjSCnQ, kDDEQUfMT_8, zY4K3sEL5lU, 9jiyee7zCXo, qQrY8ZjSCnQ, sG548miBvh0, 9up7LiqU9u4, qQrY8ZjSCnQ, GLeoET3gZWg, 2t18v4HnQxY, qQrY8ZjSCnQ, zWDJojt_KTU, Ab6ryHD_ahQ, qQrY8ZjSCnQ, COls47WPPNk, FtN8XzP8ZWg, In35ZJEGtu8, l6L2xaYRm7U, qQrY8ZjSCnQ, 9zy1SdoIpkA, nIWnC9RMW08, l6rHUHZx5Vc, xw39PKAXkdE, qQrY8ZjSCnQ, omhUqE0SANk, R5gE6vK8c7Y, 9jiyee7zCXo, pauj3PKE6L4, 9R4O3bzONQQ, omhUqE0SANk, zY4K3sEL5lU, omhUqE0SANk, 6vcXN7ko3Fk, COls47WPPNk, FmtbYlybAlo, zWDJojt_KTU, 9R4O3bzONQQ, 9jiyee7zCXo, i_r3gCZrdSc, qQrY8ZjSCnQ, omhUqE0SANk, qQrY8ZjSCnQ, 541Dq1D1_JY, 9R4O3bzONQQ, 9up7LiqU9u4, l6rHUHZx5Vc, qQrY8ZjSCnQ, sk53YzngvEs, ZtUv67HHw6g, qQrY8ZjSCnQ, l6rHUHZx5Vc, bLVLhLMoV70, COls47WPPNk, 9jiyee7zCXo, omhUqE0SANk, qQrY8ZjSCnQ, POXAC1WpDBU, qQrY8ZjSCnQ, d5NEhm7ZOyc, DUAdarwNCRs, 5YrhJMiyaPM, l6L2xaYRm7U, zWDJojt_KTU, DUAdarwNCRs, pauj3PKE6L4, sk53YzngvEs, omhUqE0SANk, Ab6ryHD_ahQ, OB9GAgiEVqk, xw39PKAXkdE, WyugIciIeq8, s4HerxySu4c, pauj3PKE6L4, 9R4O3bzONQQ, qQrY8ZjSCnQ, omhUqE0SANk, DUAdarwNCRs, 9jiyee7zCXo, qQrY8ZjSCnQ, l6L2xaYRm7U, WyugIciIeq8, GLeoET3gZWg, In35ZJEGtu8, WyugIciIeq8, pkc-AJTfNDY, jzcrUdexbNE, POXAC1WpDBU, i_r3gCZrdSc, POXAC1WpDBU, FtN8XzP8ZWg, Ab6ryHD_ahQ, qQrY8ZjSCnQ, 6vcXN7ko3Fk, VE0v1gsOMlw, GLeoET3gZWg, s4HerxySu4c, omhUqE0SANk, 9R4O3bzONQQ, 2MZR4cT7GbY, nIWnC9RMW08, d5NEhm7ZOyc, xw39PKAXkdE, Yvy_WkwkufU, d5NEhm7ZOyc, s4HerxySu4c, 9up7LiqU9u4, l6rHUHZx5Vc, WyugIciIeq8, 9zy1SdoIpkA, l6L2xaYRm7U, qQrY8ZjSCnQ, xw39PKAXkdE, i_r3gCZrdSc, jzcrUdexbNE, DUAdarwNCRs, d5NEhm7ZOyc, 5YrhJMiyaPM, qQrY8ZjSCnQ, nIWnC9RMW08, Yvy_WkwkufU, 5YrhJMiyaPM, sk53YzngvEs, qQrY8ZjSCnQ, l6L2xaYRm7U, 2t18v4HnQxY, qQrY8ZjSCnQ, l6L2xaYRm7U, DUAdarwNCRs, ZtUv67HHw6g, i_r3gCZrdSc, s4HerxySu4c, COls47WPPNk, zWDJojt_KTU, l6L2xaYRm7U, ZtUv67HHw6g, omhUqE0SANk, nIWnC9RMW08, 9up7LiqU9u4, 5YrhJMiyaPM, qQrY8ZjSCnQ, 9up7LiqU9u4, pauj3PKE6L4, VE0v1gsOMlw, GLeoET3gZWg, VE0v1gsOMlw, Ab6ryHD_ahQ, WyugIciIeq8, qQrY8ZjSCnQ, jzcrUdexbNE, l6L2xaYRm7U, POXAC1WpDBU, WyugIciIeq8, FtN8XzP8ZWg, qQrY8ZjSCnQ, GLeoET3gZWg, In35ZJEGtu8, 6vcXN7ko3Fk, GLeoET3gZWg, pauj3PKE6L4, nIWnC9RMW08, FtN8XzP8ZWg, WyugIciIeq8, DUAdarwNCRs, xw39PKAXkdE, l6L2xaYRm7U, SPxCCLU0cTs, Ab6ryHD_ahQ, xw39PKAXkdE, pauj3PKE6L4, sG548miBvh0, i_r3gCZrdSc, SSNfXY_xwxc, POXAC1WpDBU, WyugIciIeq8, qQrY8ZjSCnQ, s4HerxySu4c, 5YrhJMiyaPM, qQrY8ZjSCnQ, bLVLhLMoV70, 9jiyee7zCXo, qQrY8ZjSCnQ, Yvy_WkwkufU, POXAC1WpDBU, 2t18v4HnQxY, qQrY8ZjSCnQ, zWDJojt_KTU, Ab6ryHD_ahQ, l6L2xaYRm7U, POXAC1WpDBU, l6rHUHZx5Vc, COls47WPPNk, VE0v1gsOMlw, pkc-AJTfNDY, VE0v1gsOMlw, UFADTgwc4ew, COls47WPPNk, s4HerxySu4c, l6L2xaYRm7U, 6vcXN7ko3Fk, l6L2xaYRm7U, omhUqE0SANk, s4HerxySu4c, FmtbYlybAlo, xw39PKAXkdE, GLeoET3gZWg, qQrY8ZjSCnQ, l6L2xaYRm7U, d5NEhm7ZOyc, 9jiyee7zCXo, d5NEhm7ZOyc, qQrY8ZjSCnQ, VE0v1gsOMlw, i_r3gCZrdSc, 9R4O3bzONQQ, d5NEhm7ZOyc, 9R4O3bzONQQ, 9up7LiqU9u4, COls47WPPNk, 9jiyee7zCXo, bLVLhLMoV70, GLeoET3gZWg, zWDJojt_KTU, UFADTgwc4ew, 9up7LiqU9u4, 9R4O3bzONQQ, VE0v1gsOMlw, POXAC1WpDBU, ZtUv67HHw6g, WyugIciIeq8, d5NEhm7ZOyc, qQrY8ZjSCnQ, 9R4O3bzONQQ, s4HerxySu4c, i_r3gCZrdSc, 4geYwpsHQd8, Ab6ryHD_ahQ, omhUqE0SANk, qQrY8ZjSCnQ, l6rHUHZx5Vc, pauj3PKE6L4, omhUqE0SANk, COls47WPPNk, qQrY8ZjSCnQ, 9R4O3bzONQQ, SPxCCLU0cTs, l6rHUHZx5Vc, POXAC1WpDBU, i_r3gCZrdSc, qQrY8ZjSCnQ, SPxCCLU0cTs, l6rHUHZx5Vc, qQrY8ZjSCnQ, sk53YzngvEs, l6rHUHZx5Vc, POXAC1WpDBU, 9jiyee7zCXo, Ab6ryHD_ahQ, WyugIciIeq8, COls47WPPNk, WyugIciIeq8, GLeoET3gZWg, qQrY8ZjSCnQ, VE0v1gsOMlw, 6vcXN7ko3Fk, bLVLhLMoV70, VE0v1gsOMlw, xw39PKAXkdE, UFADTgwc4ew, qQrY8ZjSCnQ, GLeoET3gZWg, l6rHUHZx5Vc, omhUqE0SANk, xw39PKAXkdE, COls47WPPNk, 2t18v4HnQxY, l6L2xaYRm7U, SSNfXY_xwxc, GLeoET3gZWg, WyugIciIeq8, FmtbYlybAlo, qQrY8ZjSCnQ, l6L2xaYRm7U, Ab6ryHD_ahQ, pkc-AJTfNDY, xw39PKAXkdE, mL6-d5mTQuM, l6L2xaYRm7U

Source noise

Source	License
Audioset+FreeSound noise in DNS5 Challenge	CC BY 4.0
WHAM! 48kHz noise	CC BY-NC 4.0
DEMAND	CC BY-SA 3.0
TUT Urban Acoustic Scenes 2018 Mobile, Evaluation dataset	Non-Commercial
FSD50K	mixed: CC0, CC BY, CC BY-NC, CC Sampling+
SoundingEarth	CC BY 4.0
additional noise audios from Freesound	CC0

Source RIRs

Source	License
Simulated RIRs in DNS5 Challenge	CC BY 4.0
SLR28	Apache 2.0
MYRiAD	CC BY-NC-SA 4.0
BRAS	CC BY-SA 4.0
BUT_ReverbDB	Apache 2.0
voiceHome	CC BY-NC-SA 4.0
additional RIRs from Freesound	CC0

Citation:

Please cite this paper if you use this dataset in your research:

@inproceedings{URGENT-Zhang2024,
  title={{URGENT} Challenge: Universality, Robustness, and Generalizability For Speech Enhancement},
  author={Zhang, Wangyou and Scheibler, Robin and Saijo, Kohei and Cornell, Samuele and Li, Chenda and Ni, Zhaoheng and Pirklbauer, Jan and Sach, Marvin and Watanabe, Shinji and Fingscheidt, Tim and Qian, Yanmin},
  booktitle={Proc. Interspeech},
  pages={4868--4872},
  year={2024},
}