Audio & Video to Text - Whisper transcription with SRT and VTT subtitles

Send direct links to audio or video files; get the transcript, timed segments and ready SRT/VTT subtitles - from $0.008 per minute ($0.48 per hour), failed files free.

Two modes: fast for clear speech and accurate for accents, noise and names. Language is detected automatically; optional translation to English and paragraph chunks with timestamps for search and RAG. Works with files served as audio/* or video/* (for example MP3, WAV, M4A, OGG, MP4) that you own or have the right to process.

Try it now: press Start with the prefilled input (a short clip from Wikimedia Commons). It costs $0.008.

Important - what this Actor does not accept. It does not download from video or audio platforms. Links to YouTube, TikTok, Instagram, Facebook, X (Twitter), Vimeo, Spotify, Loom, SoundCloud, Twitch and similar services are refused at once with a clear error, and the refusal is free: the terms of those platforms forbid downloading their media. Use files you own or have the right to process (your own recordings, meetings, podcasts you publish, interviews you made, public-domain and openly licensed audio).

What you get per file

• text - the full transcript; segments[] - {id, start, end, text} with times in seconds
• srt and vtt - subtitle files as strings (switch off with Include SRT and VTT)
• language and language_probability (auto-detected, or the language you set); optional translation to English
• chunks[] for RAG - about N characters, cut on segment boundaries, each with start / end in seconds and chars (same idea as the chunks of our document Actor: index, text, chars, with a time range instead of pages)
• keyframes[] - for video, optional: up to N scene-change JPEG frames saved to the run's key-value store (time, key, url)
• duration_seconds, minutes_billed, words, model, mode, elapsed_seconds
• failed files come back as status: "error" with the reason and are free

Modes

• fast - Whisper base model, int8 on CPU. Quickest and cheapest. Good for clear speech.
• accurate - Whisper small model with beam search. Fewer errors with accents, noise and proper names; several times slower, so the price per minute is higher.

Neither mode does speaker separation (diarization) and neither is a substitute for a human check on legal, medical or other critical material.

Input

• Media file URLs - direct links; the server must answer with an audio/* or video/* Content-Type (an application/ogg answer is also accepted). Web pages and application/octet-stream downloads are refused - use a key-value store for those.
• Key-value store keys (+ optional store name) - files you uploaded.
• Mode, language (auto or a Whisper code such as en, de, fr, es, ru), translate to English, SRT/VTT, silence filter, chunk size and overlap, key frames and their maximum.
• Limits: 500 MB and 3 hours per file, up to 200 files per run. Longer or larger files are reported as failed and are free.

Pricing (pay per event, per minute of audio, rounded up)

• Fast: $0.008 per minute (event minute-fast)
• Accurate: $0.014 per minute (event minute-accurate)
• Every successful file is billed at least one minute. No start fee, nothing for key frames. Failed files are not charged.
• Your maximum-charge setting is respected: before each file the Actor checks that your remaining budget covers the whole file, and stops instead of exceeding it. A file is never half-processed and half-billed.

Notes and limits

• CPU only; measured on the default 4 GB: fast mode transcribes about 15 minutes of audio per minute of run time, accurate mode about 5 to 7. Files longer than 10 minutes are processed in 10-minute windows (cut at the quietest point) so memory stays flat; the language is detected on the first window. Set a run timeout that fits your longest file. With less than 4096 MB of Actor memory the limit is 60 minutes per file.
• Quality depends on audio quality. Whisper can invent words in silence or music; the silence filter (on by default) reduces this. Check language_probability for files with mixed languages.
• Only the files you provide are processed. A URL is fetched once, as a plain download, with the user agent SipkiMediaToText/1.0 (+https://apify.com/sipki/media-to-text). If the host answers 401, 403 or 429, the file is reported as failed (free): no retry, no workaround, no proxy.
• Links to private or internal addresses (private ranges, loopback, link-local, cloud metadata such as 169.254.169.254) are refused, also after redirects.
• Record and transcribe people only with their consent: recording and wiretapping laws differ by country and are your responsibility.
• Open-source components: OpenAI Whisper models and code (MIT), faster-whisper (MIT), CTranslate2 (MIT), PyAV with bundled FFmpeg libraries (PyAV BSD-3; FFmpeg LGPL/GPL, used only to decode and to take key frames), Pillow (HPND). Whisper model files: Systran/faster-whisper-base and -small (MIT).

How to use

1. Paste direct file links into Media file URLs, or upload files to a key-value store and list their keys.
2. Pick fast or accurate, the language (or auto) and whether you need subtitles or chunks.
3. Press Start. For long files set the run timeout to fit (fast mode handles about 15 minutes of audio per minute of run time). Call it from the Apify API, a schedule, an integration or the Apify MCP server.

Who uses it

• Podcasters and video teams - subtitles and transcripts of their own recordings.
• Researchers and journalists - transcribe interviews they recorded.
• Teams building search over meetings and calls - text with timestamps and chunks.

More questions

Can I paste a YouTube or TikTok link? No. Platform links are refused (free) because their terms forbid downloading media. Use files you own. Does it separate speakers? No, there is no speaker diarization. How much does one hour cost? $0.48 in fast mode, $0.84 in accurate mode.

Not affiliated

Not affiliated with, endorsed or sponsored by OpenAI, SYSTRAN, the FFmpeg project or any other third party named in this description.

Content rights

You provide the media and are responsible for having the right to process it and to use the transcript. The Actor does not check copyright or licences of your files.

Acceptable use

This Actor is a tool. You decide which files it processes and what you do with the results, and you are responsible for that use. Use it only where you have the right to access and process the media, and in line with applicable law (including recording-consent, data-protection and copyright law) and the terms of the sources. Do not use it to bypass blocks or access controls, to download from platforms that forbid it, or to place excessive load on any server. If a site owner asks us to stop, we may exclude that site. Sipki, LLC does not receive or store your results; they stay in your Apify storage.

No warranty

The Actor and its output are provided "as is", without warranty of accuracy, completeness or fitness for a purpose. Speech recognition makes errors. Verify results before relying on them. Liability of Sipki, LLC is limited as set out in the Apify Standard Actor Contract that governs your use of this Actor.

Personal data

Recordings and transcripts usually contain personal data (voices, names, health, financial or other details). For such data you are the controller and Sipki, LLC acts only as your processor under section 5 of the Apify Standard Actor Contract. Files and transcripts are processed only to produce your dataset; the Actor stores nothing outside your run's storages, and the creator does not access run data except on your support request. You need a lawful basis for your processing, the consent of the people recorded where the law requires it, and you must handle data-subject requests. To request deletion of anything related to you, open an Issue on this Actor's page.

For site owners (opt-out)

When you give the Actor a link, it downloads that single file with the user agent SipkiMediaToText/1.0 (+https://apify.com/sipki/media-to-text). It does not crawl and does not read robots.txt, because it only fetches links that a user supplies. To have your domain excluded, or to report a problem, open an Issue on this Actor's page: your domain will be added to the block list within 3 business days, and every run checks that list.