Check the Sandbox entry for the GALE Multi-Modal Example at http://uima.apache.org/sandbox.html#gale.multimodal.example We developed a typesystem that contained a basic type (org.gale.AudioSpan) that is an analog to Annotation. The begin & end features are in (float) seconds relative to the sofa data. We didn't put the audio in the CAS, but used the setSofaDataURI method to provide a URI reference and mimetype describing the format of the audio. AudioSpan and its subtypes are in the GaleSpeechTypes descriptor. I have promised to add some sample code illustrating how we used this typesystem for a task involving speech recognition and machine translation ... hopefully within the next few weeks ....!
- Burn.
