SGX 3 Release Notes | Speech Graphics Knowledge Base

SGX 3.6.2

Fixed issue causing the phonetic analysis to be mistakenly not used for short isolated words.

Maya 2022 support
An option has been added for nondeterministic processing, to get a different result (in terms of nonverbal behavior) each time you process the same input.

In the Import function of the SGX Maya plugin, audio import is now optional rather than happening automatically.
The Batch Export function of the SGX Maya plugin now accepts a list of input files as well as a directory.
Immediately adjacent markup tokens in a transcript (without time values to separate them) are now evaluated sequentially, with later markup values overriding earlier ones.
In the transcript list input file, the name of the audio at the beginning of each line, which was required to be the base name of the file, may now be a full filename or path. This should make it easier to construct the transcript list input file from existing file lists.
The transcript list input file may be a .csv file.
An error is logged when a transcript has no corresponding audio file.
Memory leaks fixed.

SGX 3.3.0

New 'extract animation' option in SGX batch processing, which extracts animation in .rts format alongside events.

Pre-roll and post-roll now applied to metadata
Muscles now guaranteed to return to neutral with 'neutral finish' option (if not transcriptionless)
Muscles now guaranteed to start from neutral, given adequate pre-roll (if not transcriptionless)
SGX Maya commands now work with python syntax. Documentation updated.
Default output folder is not overwritten. A separate unique folder is created, e.g. "output(2)".
Bug fixes:
- UE4 Player Sample Project - fix for blendshape data parsing crash
- Fix for infinite loop in rare edge case in motion synthesizer
- Transcriptionless processing now responds to 'speech_magnitude' markup

Fixes for silent and corrupted audio files
Fixed non-determinism in mood detection for transcriptionless processing
Compatible with new lightweight character file format.
Metadata
- Animation in .rts files for use in engine is now expressed in relative values (same format as SG Com)
- New SGX UE4 Player to play .rts files
- .rts export has been removed from SGX Maya plugin
- Automatically detected moods are now included in metadata

Automatic mood detection. Use 'mood=auto' in markup to utilize. Requires moods 'neutral, positive, negative' to be present in the Moods. (We will sync SG Com 3.2 to work same way.)
Markup changes
- Markup is now interpreted such that if a parameter is left unspecified in a given markup token, that parameter continues the value it had in the previous token.
- Markup tokens will not cause expression breaks unless they dictate a change in mood.
- Fix for failure to use jawmax in batch markup.
- Batch markup strings can have parentheses if desired, like transcript markup.
- Custom speech_magnitude markup parameter
New 'overwrite' option to replace output that has already been processed.
Reduction in unnecessary autocorrections.
Added automatic check for stereo input files. In case of stereo, reports error and the file is added to the failed items.
Not attempting to process files in the input directory that are not audio or text.
A new output .txt file is saved next to the .csv batch report file; the new file records the command arguments that were used for the batch.
Better handling for empty expressions or missing expressions in expression library.
Relabelled "Transcripts file" to "Batch transcript file" in SGX-GUI.
Fixed bug that caused failure for short non-speech files.
Fixed several issues in motion generation (arcing error, phase merge error, postural sustain final, postural bidirectional motion unwanted return to neutral).