This is a small program for transcribing the content of WAV files. You pass it a Whisper model, a list of WAV files, and a CSV path. It writes a CSV file containing the filenames and the transcription of the file.
I made this to automatically transcribe a rip of the WAV files from Ace Combat 5 and Ace Combat Zero that someone posted onto the internet years ago. It's close enough that you can use it to find the file containing the line you need. However, it needs manual correction for use as in-game/in-video subtitles.
To use this, you'll need the Go programming language tools
Download a ggml whisper.cpp model from https://huggingface.co/ggerganov/whisper.cpp. I found that the medium model worked pretty well, but you should experiment on your data.
Then run:
C_INCLUDE_PATH=third_party/whisper.cpp/ \
LIBRARY_PATH=third_party/whisper.cpp/ \
go run cmd/main.go \
--model $path_to_model \
--csv out.csv \
file1.wav file2.wav file3.wav
C_INCLUDE_PATH and LIBRARY_PATH should be a directory containing whisper.h and libwhisper.a, respectively. --model should be the whisper.cpp model downloaded from huggingface. --csv is the CSV file that will be created. The remaining arguments are the WAV files to transcribe.
A sample script is included in scripts.