Obsidian voice recognition with local Whisper model

readme.md

Obsidian voice recognition with local Whisper model

Install plugin Whisper for Obsidian

plugin's settings:

• API-key = sk-0
• API-URL = http://127.0.0.1:8000/inference
• language = en
• recording folder = rec
• transcription folder = rec_notes

create folders rec and rec_notes in Obsidian.

---

Install Whisper.cpp

https://github.com/ggerganov/whisper.cpp

all binaries for 1.6.2 (select your OS on the left, click "Upload binaries" section and check "Artifact download URL")
https://github.com/ggerganov/whisper.cpp/actions/runs/9251036100/job/25445724414

direct link for Whisper.cpp binaries windows-cublas (Release v1.6.2, x64, ON, ON, 12.2.0) https://github.com/ggerganov/whisper.cpp/actions/runs/9251036100/artifacts/1540277804

---

Download whisper.cpp model

from https://huggingface.co/ggerganov/whisper.cpp/tree/main and put in models folder
in my case - ggml-large-v3-q5_0.bin

---

create file start_whisper_server.bat in whisper.cpp folder and start the server

You're awesome!

win11+gpu tested
updated Jun 18 2024

start_whisper_server.bat

@echo on
cd %~dp0
.\server.exe --convert -pp -debug -l auto -m .\models\ggml-large-v3-q5_0.bin --port 8000
pause

:: use --convert if input file not wav ar 16000 -ac 1 -c:a pcm_s16le

works great! thank you

Works perfect, explained so well bro. Amazing results. Now in Obsidian it is possible to capture any voice from onboard, trancribe it, translate it, ask mistral to reprhase the conent, use styleTTS2 to make a speech again, even more with rvc inferences. All without internet, in a medium budget laptop. This is getting wild.

The api url had to be "http://127.0.0.1:8000/inference" instead of just "127.0.0.1:8000/inference" for me to get this running.