trained on mostly 15ai generated audio, the conversion are not of the best quality but it may be masked with some de-nosing and use of music/sound effects.