The DataObfuscation
task allows you to process training data audio files and labels in such a way that any sensitive or private information is hidden, so that the data can be used for model training.
Parameters | Description | Required |
---|---|---|
Type | The task name. Set to DataObfuscation . |
Yes |
Am | The acoustic model to use for processing. | Yes |
AudioAnalysisDir | The location of the audio analysis output directory. | Yes |
BeamStep | The amount to increase the beam value by on a pass failure, before attempting another pass. | |
DataList | A list of the files to use for processing. | Yes |
Diag | Whether to generate diagnostic information. | |
DiagFile | The file to write the diagnostic information to. | |
DnnFile | The DNN acoustic model file to use for processing. | Yes |
MaxBeam | The maximum beam value at which to attempt the adaptation pass. | |
MaxOtdLen | The required maximum length of each obfuscated training data file, in seconds. | |
MinBeam | The minimum beam value at which to attempt the adaptation pass. | |
OtdPath | The directory to write the obfuscated training data files to. | Yes |
OutLabExt | The label file extension. | |
OutLabPath | The directory to write label files to. By default, IDOL Speech Server writes the files to the configured temp directory. | |
Pgf | The pronunciation generation (.pgf) file included in the language pack. | Yes |
PlhExt | The file extension of the input audio feature files. | |
PlhPath | The path to the directory containing the acoustic feature (.plh) files specified in the DataList. | Yes |
RandOtd | Whether to randomize obfuscated training data. | |
TxtExt | The file extension of the input transcription files. | |
TxtPath | The path to the directory containing the transcript (.ctm) files specified in the DataList. | Yes |
WriteOutLabs | Whether to create label files. |
http://localhost:13000/action=AddTask&Type=DataObfuscation&Am=ver-ENUK-tel-6.2-8k.am&Pgf=ver-ENUK-tel-6.2.pgf&DnnFile=ver-ENUK-tel-6.2-8k.dnn&DataList=ListManager/ObfuscList&PlhPath=T:\data\PLH&TxtPath=T:\data\ObfuscTranscripts&OtdPath=T:\ObfuscTraining&AudioAnalysisDir=T:\AudioAnalysis
This action uses port 13000
to instruct Speech Server, which is located on the local machine, to produce the obfuscated training data (.otd
) file, using the ver-ENUK-tel-6.2-8k.am
acoustic model file, the ver-ENUK-tel-6.2.pgf
pronunciation generation file, and the set of data files specified in the DataList
parameter to process the data. Speech Server writes the obfuscated training data files to the ObfuscTraining
directory.
|