Support for Apple Silicon and real-time Python, and no longer any need for external shared buffers.

We increased our two-database accuracy from 76.1% to 83.3%.

Bug fix in VokaWav.h for stereo files.

We always had a function that fills the VokaturiVoice with 64-bit floating-point data. Now you can also fill the VokaturiVoice directly with 32-bit floating point data, 16-bit integer data, or 32-bit integer data, without first converting. You can now also directly send interleaved stereo data to two VokaturiVoices, in all four supported formats.

We included libraries for the Raspberry Pi.