First impression of whisper

Tried out openai/whisper

Whisper is a general-purpose speech recognition model.

Install

It also requires the command-line tool ffmpeg to be installed on your system,

# on MacOS
brew install ffmpeg

OMG. Installing one package on brew takes so long

Had some errors trying to run whisper

Traceback ...
...
[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: unable to get local issuer certificate (_ssl.c:1125)
...

Fixed the error by running following command (change 3.8 to your version of python)

/Applications/Python\ 3.8/Install\ Certificates.command

Run

Command I used for transcribing

whisper audio_en.mp3 --model base --language en
whisper audio_ko.mp3 --model base --language ko

Using any other models bigger than 'base' was way too slow for my old macbook (2015).

Output

Command line shows some sample results on the terminal. Generates transcribed files: *.vtt and *.txt

Impressions

Awesome speech recognition for English. It can pick up my voice reading stuff in English with some accent flawlessly.

However, voice memo that contains a conversation between me and my then-4yo daughter in Korean wasn't as impressive. Much more room to improve on Korean and possibly other non-English languages.

Grateful that these things are available.