Adding option to set n_threads for Whisper.cpp (#892)

CONTRIBUTING.md

@@ -94,31 +94,23 @@ Set-ExecutionPolicy Bypass -Scope Process -Force; [System.Net.ServicePointManage
 ```
 2. Install the GNU make. `choco install make`
 3. Install the ffmpeg. `choco install ffmpeg`
-4. Install Poetry, paste this info Windows PowerShell line by line. [More info](https://python-poetry.org/docs/)
+4. Install [MSYS2](https://www.msys2.org/), follow [this guide](https://sajidifti.medium.com/how-to-install-gcc-and-gdb-on-windows-using-msys2-tutorial-0fceb7e66454).
+5. Install Poetry, paste this info Windows PowerShell line by line. [More info](https://python-poetry.org/docs/)
 ```
 (Invoke-WebRequest -Uri https://install.python-poetry.org -UseBasicParsing).Content | py -
 
 [Environment]::SetEnvironmentVariable("Path", $env:Path + ";%APPDATA%\pypoetry\venv\Scripts", "User")
 
 Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser
 ```
-5. Restart Windows.
+6. Restart Windows.
 
-6. Clone the repository `git clone --recursive https://github.com/chidiwilliams/buzz.git`
-7. Enter repo folder `cd buzz`
-8. Copy `whisper.dll` from the repo backup to `buzz` folder. 
-```
-cp -r .\dll_backup\ .\buzz\
-```
+7. Clone the repository `git clone --recursive https://github.com/chidiwilliams/buzz.git`
+8. Enter repo folder `cd buzz`
 9. Activate the virtual environment `poetry shell`
 10. Install the dependencies `poetry install`
 11. Build Buzz `poetry build`
-12. Install Buzz 
-```
-$whlFile = Get-ChildItem .\dist\buzz*.whl | Select-Object -First 1
-pip install $whlFile
-```
-13. Run Buzz `python -m buzz`
+12. Run Buzz `python -m buzz`
 
 #### GPU Support
 

Makefile

@@ -26,10 +26,17 @@ else
 endif
 
 clean:
+ifeq ($(OS), Windows_NT)
+	del /f buzz\$(LIBWHISPER) 2> nul
+	del /f buzz\whisper_cpp.py 2> nul
+	rmdir /s /q whisper.cpp\build 2> nul
+	rmdir /s /q dist 2> nul
+else
 	rm -f buzz/$(LIBWHISPER)
 	rm -f buzz/whisper_cpp.py
 	rm -rf whisper.cpp/build || true
 	rm -rf dist/* || true
+endif
 
 COVERAGE_THRESHOLD := 75
 
@@ -68,9 +75,9 @@ else
 endif
 
 buzz/$(LIBWHISPER):
-ifeq ($(OS),Windows_NT)
-	cp dll_backup/whisper.dll buzz || true
-	cp dll_backup/SDL2.dll buzz || true
+ifeq ($(OS), Windows_NT)
+	cp dll_backup/whisper.dll buzz || copy dll_backup\whisper.dll buzz\whisper.dll
+	cp dll_backup/SDL2.dll buzz || copy dll_backup\SDL2.dll buzz\SDL2.dll
 else
 	cmake -S whisper.cpp -B whisper.cpp/build/ $(CMAKE_FLAGS)
 	cmake --build whisper.cpp/build --verbose
@@ -98,6 +105,7 @@ dmg_mac:
 		--app-drop-link 425 120 \
 		--codesign "$$BUZZ_CODESIGN_IDENTITY" \
 		--notarize "$$BUZZ_KEYCHAIN_NOTARY_PROFILE" \
+		--filesystem APFS \
 		"${mac_dmg_path}" \
 		"dist/dmg/"
 
@@ -188,10 +196,19 @@ translation_po:
 	sed -i.bak 's/CHARSET/UTF-8/' ${TMP_POT_FILE_PATH} && rm ${TMP_POT_FILE_PATH}.bak
 	msgmerge -U ${PO_FILE_PATH} ${TMP_POT_FILE_PATH}
 
+# On windows we can have two ways to compile locales, one for CI the other for local builds
+# Will try both and ignore errors if they fail
 translation_mo:
+ifeq ($(OS), Windows_NT)
+	-forfiles /p buzz\locale /c "cmd /c python ..\..\msgfmt.py -o @path\LC_MESSAGES\buzz.mo @path\LC_MESSAGES\buzz.po"
+	-for dir in buzz/locale/*/ ; do \
+		python msgfmt.py -o $$dir/LC_MESSAGES/buzz.mo $$dir/LC_MESSAGES/buzz.po; \
+	done
+else
 	for dir in buzz/locale/*/ ; do \
 		python msgfmt.py -o $$dir/LC_MESSAGES/buzz.mo $$dir/LC_MESSAGES/buzz.po; \
 	done
+endif
 
 lint:
 	ruff check . --fix

buzz/transcriber/whisper_cpp.py

@@ -1,3 +1,4 @@
+import os
 import ctypes
 import logging
 from typing import Union, Any, List
@@ -109,6 +110,7 @@ def whisper_cpp_params(
     params = whisper_cpp.whisper_full_default_params(
         whisper_cpp.WHISPER_SAMPLING_GREEDY
     )
+    params.n_threads = int(os.getenv("BUZZ_WHISPERCPP_N_THREADS", 4))
     params.print_realtime = print_realtime
     params.print_progress = print_progress
 

docs/docs/preferences.md

@@ -34,4 +34,34 @@ Available variables:
 
 Live transcription export can be used to integrate Buzz with other applications like OBS Studio. When enabled, live text transcripts will be exported to a text file as they get generated and translated.
 
-If AI translation is enabled for live recordings, the translated text will also be exported to the text file. Filename for the translated text will end with `.translated.txt`. 
+If AI translation is enabled for live recordings, the translated text will also be exported to the text file. Filename for the translated text will end with `.translated.txt`. 
+
+## Advanced Preferences
+
+To keep preferences section simple for new users, some more advanced preferences are settable via OS environment variables. Set the necessary environment variables in your OS before starting Buzz or create a script to set them.
+
+On MacOS and Linux crete `run_buzz.sh` with the following content:
+
+```bash
+#!/bin/bash
+export VARIABLE=value
+export SOME_OTHER_VARIABLE=some_other_value
+buzz
+```
+
+On Windows crete `run_buzz.bat` with the following content:
+
+```bat
+@echo off
+set VARIABLE=value
+set SOME_OTHER_VARIABLE=some_other_value
+"C:\Program Files (x86)\Buzz\Buzz.exe"
+```
+
+### Available variables
+
+**BUZZ_WHISPERCPP_N_THREADS** - Number of threads to use for Whisper.cpp model. Default is `4`. Available from `v1.0.2`. 
+
+On a laptop with 16 threads setting `BUZZ_WHISPERCPP_N_THREADS=8` leads to some 15% speedup in transcription time. 
+Increasing number of threads even more will lead in slower transcription time as results from parallel threads has to be 
+combined to produce the final answer.