Applies to older version SpeechServers (Dragon v5/6 or IBM v8+) only. Older version included Standalone desktop client edition that is no longer available.
Corrective adaptation is provided by SpeechTrainer. Training is handled through the automatic activity of CSUSA_Train_X where X is the engine type. The activity must exist on a workflow for processing.
Corrective adaptation is done by transcribing the audio, selecting differences with the transcribed output and the provided verbatim text, and applying corrections. The process is repeated until appropriate conditions are met (target accuracy, unable to correct any further, maximum number of cycles).
The source audio file can be in any supported acWAVE format (.wav, .vox, .mp3, .wma) and will be converted to the engine's requirements before training. To insure maximum accuracy, be sure to use a high-quality recording (low bit rates can cause more inaccuracies).
The source verbatim text file must be in a .txt format. There is the option to use (instead of the verbatim text file) a training .trn file generated by SpeechMax. This requires that the original transcription is processed through SaveSession and not TransWaveX.
If the job does not have an assigned engine user ID (based on Author or Job UDA settings), the engine will automatically create a new user (based on default settings) for processing. This new user will be assigned to the job's author.
When training starts, the user is locked in the repository, preventing other servers from updating the same user. Once complete, the trained user is uploaded to the repository and released for further training. Once uploaded, other servers will use the newly trained user for future activities.
The activity will complete normally if no errors occur, or will complete with a condition of ERROR in the event that an error occurred.
To perform training, do the following:
1. Create a job in Command! on a workflow that has the CSUSA_Train_X activity and assign (or create) the required audio file. Also assign (or create) the verbatim text file.
2. Be sure to have appropriate UDA settings according to SpeechParameters.