The original project by Jared Szechy written in 2014 is https://github.com/szechyjs/dsd
DSD is able to decode several digital voice formats from discriminator tap audio and synthesize the decoded speech. Speech synthesis requires mbelib, which is a separate package. The DSD Wiki has lots of additional information about DSD including build instructions and answers to the most frequently asked questions.
It is available here: https://github.com/szechyjs/dsd/wiki
The supporting mbelib is available here: https://github.com/szechyjs/mbelib mbelib supports the 7200x4400 bit/s codec used in P25 Phase 1, the 7100x4400 bit/s codec used in ProVoice and the "Half Rate" 3600x2250 bit/s vocoder used in various radio systems. It will decode the AMBE Codec but not encode.