Video to text: every way to convert a video into text

Two jobs, in order: pull the audio out of the video container, then transcribe it. Online converters (Otter, VEED, Happy Scribe, Klipa) do both in one upload — metered by minutes, credits, or watermarks. Locally, ffmpeg does the extraction (ffmpeg -i input.mp4 -vn -ac 1 -ar 16000 -c:a pcm_s16le audio.wav) and a Whisper-class engine does the transcription (whisper-cli -m ggml-large-v3-turbo.bin -f audio.wav -l en -osrt -otxt) — free, unlimited, nothing uploads. A 30-minute MP4 converts in a couple of minutes on a GPU.

The free tiers are real but capped, and each cap is a different trap: Otter allows 300 minutes a month but only 3 file imports for the lifetime of the account; Klipa caps free files at 10 minutes; Transcriptly gives 10 credits a month (a 30-minute local file eats 30 of them); VEED shows the transcript free but gates TXT/SRT/VTT downloads to paid; Happy Scribe's free tier is a 10-minute trial; Sonix gives a 30-minute one-time trial. The genuinely free converter is the local pipeline: ffmpeg + whisper-cli on your own machine, unlimited, no account.

Two commands after a one-time setup (install ffmpeg and whisper.cpp). First extract the audio: ffmpeg -i input.mp4 -vn -ac 1 -ar 16000 -c:a pcm_s16le audio.wav — that writes the 16 kHz mono WAV speech models expect. Then transcribe: whisper-cli -m ggml-large-v3-turbo.bin -f audio.wav -l en -osrt -otxt produces both the plain-text transcript and the timed SRT from one pass. Prefer an app over the terminal: VocalFuse wraps the same pipeline on Windows — drop the MP4 in, get TXT/SRT/VTT out, from $5/mo with a free tier.

Every serious converter runs a Whisper-class engine, so accuracy converges on clean English audio — roughly 92-97% word accuracy, with proper nouns the main error class. Differences show up at the edges: ElevenLabs Scribe leads on non-English and audio-event tagging, Whisper large-v3 stays the strongest open model for jargon-dense footage, and NVIDIA Parakeet posts 6.34% WER vs Whisper large-v3's 7.44% on the Open ASR Leaderboard. Vendor claims of "99.9%" are marketing on self-selected audio — test any engine on your own files.

They are different jobs. A transcript is prose for reading, quoting, show notes, and LLM input — publishing it on the page is also what makes the video indexable by search engines. Subtitles (SRT/VTT) are timed cue files for a video editor or a player: SRT uploads to YouTube's caption editor, VTT is the HTML5 web format. Most engines produce both from one pass. Watch converter fine print: several free tiers let you download TXT but gate SRT/VTT to paid — you get text you cannot load back into your editor.

On a GPU with a large-v3-class model, about 5-10 minutes plus the one-command audio extraction (seconds). On CPU, roughly 0.3x real time — about 20-30 minutes for an hour of footage, faster with turbo or small models. Cloud engines advertise similar processing times but add upload time (an hour-long MP4 is hundreds of MB to GB) and queue time on top. Batch folders convert locally overnight, unattended, at zero per-hour cost.

Every cloud converter works by upload — that is the architecture: your MP4 goes to their servers, their engine transcribes it, you download text. SOC 2 certifications and deletion policies govern what happens after the upload; they do not remove it. For unreleased footage, internal training video, client work, or anything confidential, the local pipeline is the only route where no server ever receives the file: the extraction and transcription both run on your machine, and nothing leaves it.

Yes, and the route depends on the video. If the video has a caption track, you do not need conversion at all — copy the Show transcript panel (in the expanded description since October 2025) or pull the track with yt-dlp, free and instant. If it has no captions, download the audio with yt-dlp (yt-dlp -x --audio-format wav) and transcribe it locally — that is real speech recognition, the same pipeline as any other video file. The full route-by-route guide is at /youtube-to-text.