34 datasets found

Synthetic test set for DNS. Contains 300 far-end and near-end signal pairs with SER of 0, 3.5, 7 dB.

Real-recorded test set for DNS. Consists of 800 real-world recordings including various background noises at different SNR, target levels, acoustic conditions, and languages.

Real-recorded test set for AEC. Consists of 800 real-world recordings including three talking scenarios: doubletalk, farend-singletalk and nearend-singletalk.

Synthetic test set for ablation study on AEC. Contains 500 hours of audio samples for training and 8 hours for validation. The far-end, near-end and the noise signals are all...

Segan: Speech enhancement generative adversarial network.

The DEMAND and QUT-TIMIT datasets were used to create the corrupted speech input tracks according to the following signal-to-noise ratios (SNRs): 0 and 5 dB.

The Voice Bank corpus dataset was used as the ground-truth reference tracks. Additive noise from the DEMAND and QUT-TIMIT datasets were used to create the corrupted speech input...

The QUT-NOISE-TIMIT dataset is a dataset for speech enhancement. It consists of clean speech and noise.

The dataset used for training and validation of the proposed deep binaural MFMVDR filter, comprising simulated binaural room impulse responses and clean speech and noise.

The dataset used for evaluation of the proposed deep binaural MFMVDR filter, comprising measured binaural room impulse responses and clean speech and noise.

The dataset used for training and validation of the proposed deep binaural MFMVDR filter, comprising simulated binaural room impulse responses and clean speech and noise.

Noisy speech database for training speech enhancement algorithms and TTS models.

The MHINT corpus is a Mandarin Chinese speech corpus used for speech recognition and speech enhancement. It contains 480 utterances of 10 speakers, each with 10 seconds of speech.

The proposed self-supervised speech quality estimator trained only on clean speech.

The database in [28] which was used to evaluate SEGAN in [14].

Speech enhancement dataset

Diffusion probabilistic model for speech enhancement

The dataset used in this paper is a collection of phonetically and phonologically local allophonic distribution in English, where voiceless stops surface as aspirated...

The TIMIT corpus is a widely used benchmark for speech recognition tasks. It contains 3,696 training utterances from 462 speakers, excluding the SA sentences. The core test set...

The LRS2 dataset consists of 48,164 video clips from outdoor shows on BBC television. Each video is accompanied by an audio corresponding to a sentence with up to 100 characters.