Roger voice release for Blizzard 2010
These data are available only to registered participants in the Blizzard Challenge 2010.Speech data
The 16kHz speech data are identical to those in the ARCTIC subset of the corpus used in the 2008 and 2009 challenges, so if you obtained the data through your participation in 2008 or 2009, you do not need to complete this license form or download the data again. If you want the 48kHz files, simply complete the license again. This data is released under a license for non-commercial use only. To download, first read and accept the license. Once we have received your license, we will email you a password.
Downloads
The following files are available to download from here:- roger_arctic_blizzard_release_2010_16k.tar.bz2 - the ARCTIC subset of the roger corpus at 16kHz sampling rate
- roger_arctic_blizzard_release_2010_48k_wavs_only.tar.bz2 - the ARCTIC subset of the roger corpus at 48kHz sampling rate (wav files only - please download the 16k file to obtain transcriptions etc)
- NEW roger-out-sync-data.tar.gz corrected versions of a few waveforms from the 48kHz version which were not correctly synchronised with the labels
Labels
Two sets of labels are provided:- standard Festival utterances, created at the University of Edinburgh using the multisyn voice building tools - these are identical to those provided in 2009
- hand-corrected phone labels (based on the original Festival utterances above) and hand-annotated prosodic labels, kindly provided by iFLYTEK.
Downloads
The following files are available to download from here:- original_festival_utts_for_roger_arctic.tar.bz2 - automatically produced labels
- iFLYTEK_labels_for_roger_arctic.tar.bz2 - hand-corrected labels
Development data, tools, benchmark voices
Several items are available for use by participants during development of their voice. These include the 2009 test sentences synthesised by this year's benchmark voices. Since these items depend on the Unisyn lexicon, they are only available to participants who have agreed to the Unisyn lexicon license, as for the labels above.Downloads
The following files are available to download from here:- roger-arctic-benchmark-voice-hts-48k-original_labels-ver1.tar.gz - the HTS benchmark voice trained on the 48kHz version of the data using the automatically produced labels, including the 2009 test set synthesised using this voice plus everything you need to build your own version of this voice
- roger-arctic-benchmark-voice-hts-48k-iflytek_labels-ver1.tar.gz - the HTS benchmark voice trained on the 48kHz version of the data using the hand-corrected iFLYTEK labels, including the 2009 test set synthesised using this voice plus everything you need to build your own version of this voice (NOT YET AVAILABLE)
- roger-arctic-benchmark-voice-festival-16k-iflytek_labels-ver1.tar.gz - the 2009 test set synthesised by the Festival benchmark voice built from the 16kHz version of the data and the automatically-produced labels (NOT YET AVAILABLE)
Test sentences
Start hereContact Simon King for more details.