Recently contributed a patch to FluidAudio that sped up Parakeet V2 and V3 to 320x and 282x faster than real time, respectively:
replyhttps://github.com/FluidInference/FluidAudio/pull/507
That means one hour of audio transcribed in 11.25 and 12.75 seconds.
The Inscribe post doesn't give a speed factor for SpeechAnalyzer. However, this Argmax blog post reports 70:
https://www.argmaxinc.com/blog/apple-and-argmax
Based on that, FluidAudio is ~4.6x and ~4.0x faster.
Just tried test using yap on a single ~1hr mp3: yap/Speechanalyzer is about 50% slower than fluidaudio on M1. yap interface is nicer though.
reply