Transcribr

Bulk Audio & Video Transcriber

A 1-click Pinokio launcher that transcribes audio in bulk using faster-whisper — from one or many YouTube videos (or entire playlists) and/or your own uploaded audio/video files.

What it does

• Paste any number of YouTube video or playlist URLs (one per line).
• Or upload your own audio/video files to transcribe them directly.
• Downloads the best audio with yt-dlp (bundled ffmpeg, no system install).
• Transcribes each item with a Whisper model (tiny → large-v3).
• Runs on GPU (NVIDIA, automatic) or CPU.
• Exports each transcript as txt, srt, vtt, or json — a single transcript is shown inline and downloadable as-is, several are bundled into a .zip.

How to use

1. Click Install (installs dependencies; Whisper models download on first run).
2. Click Start, then open the Web UI.
3. Paste YouTube URLs and/or upload audio/video files, then choose a model size, device, language, and output format.
4. Click Transcribe. Results appear in the table; one transcript is shown in the text box, several arrive as a .zip download.

Transcripts are also saved under app/transcripts/run_<timestamp>/.

Options

Option	Description
Model size	tiny, base, small, medium, large-v3 (bigger = better/slower)
Device	auto (GPU if available), cpu, cuda
Language	Blank = auto-detect, or an ISO code such as en, fr, es
Output format	txt, srt, vtt, json

Programmatic access

The transcription engine is plain faster-whisper + yt-dlp, so you can script the same pipeline directly.

Python

import yt_dlp
from faster_whisper import WhisperModel

url = "https://www.youtube.com/watch?v=VIDEO_ID"
opts = {"format": "bestaudio/best", "outtmpl": "audio.%(ext)s", "noplaylist": True}
with yt_dlp.YoutubeDL(opts) as ydl:
    info = ydl.extract_info(url, download=True)
    # prepare_filename() reports the pre-processing name; the finished file is
    # recorded under requested_downloads.
    audio = info["requested_downloads"][0]["filepath"]

model = WhisperModel("small", device="auto", compute_type="int8")
segments, info = model.transcribe(audio)
print("".join(seg.text for seg in segments))

JavaScript (Node)

There is no JavaScript transcription engine here — shell out to the same Python tools that this app installs into its env virtualenv.

import { spawnSync } from "node:child_process";

// Download audio
spawnSync("yt-dlp", ["-f", "bestaudio/best", "-o", "audio.%(ext)s",
  "https://www.youtube.com/watch?v=VIDEO_ID"], { stdio: "inherit" });

// Transcribe with faster-whisper
spawnSync("python", ["-c", `
from faster_whisper import WhisperModel
segments, info = WhisperModel("small", compute_type="int8").transcribe("audio.webm")
print("".join(s.text for s in segments))
`], { stdio: "inherit" });

Curl

yt-dlp and faster-whisper are local tools, not an HTTP API. To fetch a single video's audio without this app:

yt-dlp -f bestaudio/best -o "audio.%(ext)s" "https://www.youtube.com/watch?v=VIDEO_ID"

Requirements

• ~1–3 GB disk for Whisper models (downloaded on demand).
• NVIDIA GPU optional; CPU works for smaller models.