SCC

Wat erin zit

A synthesized dataset consisting of 135,000 wave files in two separate sets of train, dev and test splits. Each wave file has a certain voice uttering a certain sentence, with certain channel properties (noise, reverb, and filtering). For both sets of splits, there are 100 different speakers, 100 different sentences, and 100 different channel settings. The dataset can be used for signal processing classification tasks of speaker, content and channel attributes. It was constructed for research into automatic speaker verification, see https://github.com/NetherlandsForensicInstitute/interpretability_in_speaker_verification. The code for this project can be used to generate a dataset like this one (depending on random seeds and computer architecture, it might be exactly the same or have different examples). The dataset as provided here can be used independently of the project. The archive is set up as follows: In the root are 18 files with the provided splits, consisting of lines of a class label and a filepath. There are also 2 files which show which files from VOiCES and LJ Speech were used in synthesis. The 'wav' folder has the main wave files of the dataset, …

Bestanden en diensten

• DataCite REST API (JSON)api.datacite.org
  Openen