SGX Batch Processing

SGX is designed to efficiently produce quality facial animation for large amounts of audio. To facilitate this, we provide batch processing. Batch processing is performed by SGX Producer, a command-line executable for Windows. There is also a user-friendly batch processing interface in SGX Director, which launches SGX Producer batches and/or prints the commands by which they can be launched.

There are two modes of batch processing:

• processing new events

• processing existing events

Processing new events

In this mode, new audio and transcripts are specified as input. To prepare input for batch processing, see Batch Input Formats. For each input audio file a new event file will be created, with the same base name as the audio file. The example below shows a set of input files (left), and the corresponding, fully processed event files (right).

Processing existing events

When processing a batch of existing events, a directory containing the events is specified as input. The events files are processed in place – meaning no new files are created; the existing ones are overwritten. There are two reasons why you might want to process a batch of existing events:

• To reprocess them

• To process them for the first time (for example because they were constructed in SGX Director by loading audio and text files but were not processed).

Batch reports

In addition to producing or updating event files, the batch process also produces two files summarizing the batch. These will be found in the same directory as the processed event files:

where TIMESTAMP is in the format

yyyymmddThhmmss

For example:

20230929T164937

The processing report contains one row for each event, with the following comma-separated values:

name,path,processing_status,transcript_status,phonetic_analysis_score,pre_roll,post_roll

The following table provides descriptions of the reported properties for each event:

Below is an example batch report .csv file, viewed as a spreadsheet: