RATS-Channel-A-Speech-Data

We create a public repository for Robust Automatic Transcription of Speech (RATS) Channel-A Speech Data, which is a chargeable dataset under LDC (official website). Here we release its Log-Mel Fbank features and several raw wavform listening samples.

Log-Mel Fbank features

Google Drive Address: https://drive.google.com/drive/folders/118JBm0B5txiS7q09SX_xlbf2glbad0jA?usp=sharing

Where you will find a README.md for instrcutions.

Raw Waveform Listening Samples

Here we select one raw wavform sample from train set, valid set and test set, respectively. And we also present their parallel clean samples to show how noisy the RATS Channel-A corpus is.

(Note: since Github README.md does not support .wav format, we transfer the raw .wav files to .mp4 format for presentation here, one can turn to ./samples for raw files)

(1) Train Sample

a. noisy speech:

train_noisy.mp4

b. clean speech:

train_clean.mp4

(2) Valid Sample

a. noisy speech:

valid_noisy.mp4

b. clean speech:

valid_clean.mp4

(3) Test Sample

a. noisy speech:

test_noisy.mp4

b. clean speech:

test_clean.mp4

References

[1] D. Graff, K. Walker, S. M. Strassel, X. Ma, K. Jones, and A. Sawyer. “The rats collection: Supporting hlt researchwith degraded audio data.,” in LREC.Citeseer, 2014.