vds 23.3.9

ValidDataSet - TTS Lj Speech Dataset Validator

ValidDataSet

• About
• Plugins
• Installation
• Usage

About [ Menu ]

ValidDataSet was created to help validate datasets created based on the Lj Speech Dataset (for Tacotron, Flowtron, Waveglow, or RadTTS).

VDS works based on plugins (which can be dynamically added by the user in the future).

Descriptions of current plugins can be found in the Plugins section.

Plugins [ Menu ]

Below is a list of currently used plugins (new ones will be added over time).

ID	Name	Version	Description
F001	WavsTranscriptionChecker	23.3.9	Check if all files have been added to the transcription files
F002	WavPropertiesChecker	23.3.9	Check if all files are mono, 22050 Hz with length between 2 and 10 seconds
T001	DatasetStructureChecker	23.3.9	Check if the "wavs" folder and transcription files exist in the dataset
T002	EmptyLineChecker	23.3.9	Check if there are empty lines in the transcriptions
T003	FilesInTranscriptionChecker	23.3.9	Check if all files added to transcription exist
T004	ExistingWavFileTranscriptionChecker	23.3.9	Check if all files added to transcription have a transcription
T005	PunctuationMarksChecker	23.3.9	Check if all transcriptions end with punctuation marks: ".", "?" or "!"
T006	PunctuationMarksChecker	23.3.9	Check if all lines have the same number of PIPE characters
T007	DuplicatedTranscriptionChecker	23.3.9	Check if there are any duplicate paths to WAV files in the transcriptions

Installation [ Menu ]

To install ValidDataSet, use the following command:

pip install vds

Usage [ Menu ]

Command in Linux: vds or vds-win

Command in Windows: vds-win

List of parameters supported by VDS:

 -v, --verbose                    Print additional information
 -o, --output                     Save output to file

     --plugins.list               List plugins
     --plugins.disable            List of plugins to disable like: F001,T002,T006

     --args.path                  Path to dataset
     --args.files                 Set transcription file names like: train.txt,val.txt
     --args.dir-name              wavs folder name (default: wavs)
     --args.sample-rate           Set sample rate (default: 22050)
     --args.number-of-channels    Set number of channels (default: 1 [mono])
     --args.min-duration          Set minimum duration in miliseconds (1000 ms = 1 second)
     --args.max-duration          Set maximum duration in miliseconds (1000 ms = 1 second)
     --args.number-of-pipes       Set number of pipes (|) (default: 1)

Sample commands and their description:

List all plugins:

vds --plugins.list

Run VDS with all plugins without additional information:

vds --args.path /media/username/Disk/Dataset_name/

Run VDS with all plugins with additional information:

vds --args.path /media/username/Disk/Dataset_name/ -v

Run VDS without plugins F001,T002,T006 with additional information:

vds --args.path /media/username/Disk/Dataset_name/ --plugins.disable F001,T002,T006 -v

Run VDS without plugins F001,T002,T006 with own transcription names and with additional information:

vds --args.path /media/username/Disk/Dataset_name/ --plugins.disable F001,T002,T006 --args.files train.txt,val.txt -v

Run VDS and print files which are longer than 20 seconds, shorter than 2 seconds and not in mono:

vds --args.path /media/username/Disk/Dataset_name/ --args.min-duration 2000 --args.max-duration 20000 --args.number-of-channels 2 -v