faspie / whisper-workflow Goto Github PK

Purpose of this project is a small workflow script for an automated transcription of voice recorded on a speech recorder.

The script consists of a few parts

1.) Check if the voice recorder is available

2.) Copy / convert files (supported formats: MP3, DSS, DSS2 is unfortunately not supported by ffmpeg)

3.) Call API

4.) Send an email with the transcript

The script is designed for use on a raspberry pi but can also be used in many different other scenarios.

GETTING STARTED

Prerequisites on debian based systems

1.) apt-get install ffmpeg mailutils jq

2.) Configure mail system for sending mails: dpkg-reconfigure exim4-config

3.) Move script to a directory of your choice (e. g. /etc)

4.) Install it as a service

cd /lib/systemd/system/ nano workflow.service

Content of workflow.service [Unit] Description=Workflow for DPM After=multi-user.target

[Service] Type=simple ExecStart=/etc/workflow.sh Restart=on-abort

[Install] WantedBy=multi-user.target

5.) chmod 644 /lib/systemd/system/workflow.service

6.) chmod +x /etc/workflow.sh

7.) edit workflow.sh

• change recipient's mail address
• change vendor and product id of your usb device (call lsusb)
• edit function "whisper" which calls the whisper api

8.) systemctl daemon-reload && systemctl enable workflow.service && systemctl start workflow.service

API

I recommend to use the script with two local APIs which are available on github

https://github.com/morioka/tiny-openai-whisper-api and https://github.com/pluja/whishper