Repository navigation
Getting Started
= edited this page Feb 10, 2026
·
1 revision
[dependencies]
transcriber = "0.1"You also need ffmpeg on your system for audio decoding:
# Debian/Ubuntu
apt install ffmpeg
# macOS
brew install ffmpeg
# Arch
pacman -S ffmpegIf you want to transcribe URLs (YouTube etc.), install yt-dlp too:
pip install yt-dlp#[tokio::main]
async fn main() -> Result<(), transcriber::Error> {
// From a URL — downloads, decodes, transcribes
let transcript = transcriber::transcribe("https://youtube.com/watch?v=...").await?;
println!("{}", transcript.text());
// From a local file
let transcript = transcriber::transcribe_file("recording.mp3").await?;
println!("{}", transcript.to_srt());
Ok(())
}The first run will download the whisper model (~2.9 GB for large-v3) from HuggingFace. It's cached in ~/.cache/transcriber/models/ so subsequent runs are fast.
If you're not sure, start with the default (large-v3) — it's the most accurate. If it's too slow for your hardware, drop down to small or base:
use transcriber::{TranscribeOptions, Model};
let opts = TranscribeOptions::new()
.model(Model::Small);
let transcript = transcriber::transcribe_with_options(url, &opts).await?;See the Models page for a full comparison.
By default, whisper auto-detects the language. If you know what it is, setting it explicitly is faster and more reliable:
let opts = TranscribeOptions::new()
.language("de")?; // GermanYou can use short codes (de, en, fr, ja) or full names (german, english, french, japanese).