Neural audio models in C++20. Text-to-speech, speech-to-text, speaker diarization, the RAVE audio autoencoder, keyword spotting, and in-tree English G2P.