Transcripts for search
Search inside your books needs transcripts: the words with their times. A computer makes them with Whisper, free and offline, in a few minutes to half an hour per hour of audio, depending on the computer.
Each transcript is a .vtt file next to its audio file, with the same name: 01.mp3 → 01.vtt. Keep them where the book is (a connected folder, a WebDAV or web-folder server, or the book's folder on Audiobookshelf shared to Veda Audio Player that way) and Veda Audio Player picks them up when it scans.
Mac
- In Terminal (with Homebrew): brew install whisper.cpp ffmpeg
- Download the model ggml-large-v3-turbo.bin from huggingface.co/ggerganov/whisper.cpp
- For each file: ffmpeg -i 01.mp3 -ar 16000 -ac 1 01.wav
- Then: whisper-cli -m ggml-large-v3-turbo.bin -l auto -ovtt -of 01 -f 01.wav
Windows
- Download whisper-bin-x64.zip from github.com/ggml-org/whisper.cpp/releases and unzip it
- Install ffmpeg: winget install ffmpeg
- Run the same two commands in PowerShell (whisper-cli.exe)
Linux
- Install ffmpeg, then whisper.cpp (whisper-bin-ubuntu-x64 from the same releases page) or faster-whisper: pip install faster-whisper
Tips
- -l en or -l ru is faster and steadier than auto when you know the language.
- --prompt "Krishna, Bhagavad-gita, Prabhupada" helps Whisper spell the names you use.
- Leave out kirtans and music: their text is noise (they start off in Settings › Search).