I think I figured it out. When Hal was switching between brains it destroyed an instance of an SQLite object (as it should), but for some reason it wasn't being recreated correctly and was losing a pointer to the active database. I fixed it and believe the issue is resolved now, but I'm still wondering where this bug came from. I don't recall touching this code for 4 years so I'm still baffled why Hal 6.0 and 6.1 worked fine and 6.2 didn't. I'm going to try digging through some source code backups to see if I can figure that mystery out as I'm still worried there might be something else going on. But in the meantime I'm in the process of uploading the updated version Hal 6.2 RC2. Please help me test Hal's database functions thouroughly especially when switching between brains (both in the brain editor and hal assistant).
Also I fixed the Cepstral bug that was mentioned earlier in this thread. Hal 6.2 includes a new version of middleware made by another company that is used for Hal's speech synthesis and speech recognition. This new version added native support for Cepstrals Swifttalk interface. Cepstral supports both SAPI 5 and Swifttalk and Hal's options showed both interfaces (and incorrectly labeled Swifttalk as SAPI 5). However, neither Haptek nor MSAgent support Swifttalk so if this voice was chosen strange things happened. I modified Hal to hide Swifttalk voices and only show SAPI 4 and SAPI 5 voices, so that will fix that issue.
