Same here, could you let me know if you found a successful formula @Sparks ? How does it actually work in practice ? Chunks of audio are sent every X seconds to a model to decide whether it's a voicemail or not ?