"A scalable noisy speech dataset and online subjective test framework," in Interspeech, 2019. Reddy, Ebrahim Beyrami, Jamie Pool, Ross Cutler, Sriram Srinivasan, Johannes Gehrke. In addition we implemented a subjective testing method according to ITU-T P.835 which allows to rate the speech signal, background noise, and the overall quality.įurther details of this dataset can be found in our Interspeech 2019 paper:Ĭhandan K. We implemented an absolute category rating (ACR) application according to ITU-T P.800. We provide html code for building two Human Intelligence Task (HIT) crowdsourcing applications to allow users to rate the noisy audio clips.We also provide test set that is different from training set to evaluate the developed models.This dataset will immensely help researchers and practitioners in accademia and industry to develop better models.This dataset will continue to grow in size as we encourage researchers and practitioners to contribute to this dataset by adding more clean speech and noise clips.The SNR conditions and the number of hours of data required can be configured depending on the application requirements.We provide the recipe to mix clean speech and noise at various signal to noise ratio (SNR) conditions to generate large noisy speech dataset.But it can be used for other audio and speech applications. The main application of this dataset is to train Deep Neural Network (DNN) models to suppress background noise.
0 kommentar(er)
