Any tips to speed up quantized Whisper inference on Android?

Hello, running q80 quantized whisper on Android (Pixel 7) is taking around 15 seconds for 5 seconds of audio. Is there any way to speed this up that I might not be aware of or is it just because candle isn't as optimized as something like whisper.cpp yet? whisper.cpp took around 3 seconds or less if I remember correctly. Although this was with a q40 model. Thanks.

Start by reproducing the reported q80 Whisper inference benchmark on a Pixel 7, using the issue's 5-second audio case and comparing the roughly 15-second result with the cited whisper.cpp comparison. Profile the Candle Android inference path to identify the limiting component; done means a measured, documented improvement or a clear explanation of the remaining gap.

Written by the indexing model from the issue text.