My occasional voice to text uses whisper on an Intel machine. That can run on a Pi with 2 GB. It might run faster with an extra chip but it runs in the background while I am doing other things. If the Hailo chip made whisper faster, I would speed check the difference.
To run whisper on a Pi, people use a small or light model. Those models are not accurate. I use the full model which will most likely require an 8 GB Pi 5. That would be a good test of the speed gain from the extra chip when whisper supports the use of the chip.
Whisper is running now on my Intel chip at about 5 GHz. Only the medium model, not the large model. The whisper process is using 4.9 GB and total usage is 7.9 GB. The large model requires a minimum of an extra GB.
To run whisper on a Pi, people use a small or light model. Those models are not accurate. I use the full model which will most likely require an 8 GB Pi 5. That would be a good test of the speed gain from the extra chip when whisper supports the use of the chip.
Whisper is running now on my Intel chip at about 5 GHz. Only the medium model, not the large model. The whisper process is using 4.9 GB and total usage is 7.9 GB. The large model requires a minimum of an extra GB.
Statistics: Posted by peterlite — Fri Jun 07, 2024 11:36 pm