Audio & Video Transcription - Speech to Text, Whisper, SRT

Turn your own audio and video files and public podcast RSS feeds into text, timestamped segments, SRT and VTT subtitles. Works in 90+ languages, with automatic language detection. You pay only for audio that was transcribed successfully.

AI disclosure. Transcripts are produced automatically by an AI speech-recognition model: faster-whisper, an open-source (MIT-licensed) re-implementation of OpenAI's open-source Whisper model, running on Apify's servers. No human checks the output. AI transcripts can contain errors, especially with names, numbers, accents, overlapping speakers, music or poor audio. Review the transcript before you rely on it for anything important.

Why this one

• Everything you need from one run: the full transcript, sentence-level timestamps and ready-to-use SRT and VTT subtitle files, per file.
• Files and podcasts: your own audio/video links or a public podcast RSS feed, 90+ languages with automatic language detection.
• You pay only for audio that was transcribed. Unreachable URLs, non-audio files, too-long files and files with no speech are free and listed in ERRORS; files above your maximum cost per run are skipped, not charged.
• Two models: Standard (Whisper base) for fast, low-cost transcripts and Accurate (Whisper small) when the text matters more. Output is automatic and can contain errors, so review it.

Quick start: example input, output and cost

Speech to text for audio, video and podcast RSS: transcript, timestamps, SRT and VTT subtitles, 90+ languages. Pay per audio minute, failed files free.

Example input

{
  "fileUrls": [
    "https://upload.wikimedia.org/wikipedia/commons/8/8c/The_Eagle.ogg"
  ]
}

Example file: "The Eagle" (Alfred, Lord Tennyson, 1851) read by Neil Carmichael, Wikimedia Commons, licensed CC BY-SA 4.0. 16 seconds, billed as 1 minute.

Cost example: A 60-minute podcast episode x $0.00597 = $0.36 (accurate model: $1.18).

What you get per file

Field	Description
text	The full transcript
segments	Sentence-level segments with start / end times in seconds
srt, vtt	Ready-to-use subtitle files (as text)
language, languageProbability	Detected (or forced) language
durationSec, billedMinutes	Audio length and the minutes you were charged for
episodeTitle, episodePublished, feedUrl	For podcast episodes
model, engine	Which open-source model produced the transcript

Inputs that fail (unreachable URL, not an audio file, too long, no speech) are not charged. They are listed in the ERRORS record of the run's key-value store, with the reason.

Pricing (pay per event)

Event	Price	When it is charged
audio-minute (Standard, Whisper base)	$0.00597 per audio minute	After the transcript of a file has been saved
audio-minute-accurate (Accurate, Whisper small)	$0.0197 per audio minute	After the transcript of a file has been saved

Minutes are counted per file and rounded up (a 61-second file = 2 minutes). Example: a 45-minute podcast episode costs $0.27 (Standard) or $0.89 (Accurate). If you set a maximum cost per run, files that would exceed it are skipped and not charged.

Input examples

Files (direct links to files you may process):

{ "fileUrls": ["https://upload.wikimedia.org/wikipedia/commons/8/8c/The_Eagle.ogg", "https://upload.wikimedia.org/wikipedia/commons/1/16/President_Kennedy%27s_Speech_at_Rice_University.ogv"], "language": "en" }

Podcast RSS (newest 3 episodes of a NASA podcast, English, accurate model):

{ "podcastFeedUrls": ["https://www.nasa.gov/feeds/podcasts/houston-we-have-a-podcast"], "maxEpisodesPerFeed": 3, "language": "en", "model": "small" }

Which model should I choose?

• Standard (base) is fast and cheap and good for clear English speech.
• Accurate (small) makes noticeably fewer errors for Dutch, German and other non-English audio, and for noisy recordings. Our own test results are in the "Accuracy" section below.

Accuracy (our measurement, word error rate, lower is better)

Language (FLEURS samples, ~2 min each)	Standard (base)	Accurate (small)
English	18%	10%
Dutch	50%	24%
German	20%	11%

Measured on 9 Oct 2026 with the public FLEURS benchmark by Google (CC-BY-4.0): read Wikipedia sentences with many names and numbers, so these figures are on the harsh side. Clear conversational English usually comes out better. For Dutch and other non-English audio, use Accurate.

Limits and what is not supported

• Up to 200 files/feeds per run; up to 2 GB and 240 minutes per file by default (maxFileDurationMinutes, max 600).
• Not supported: page links of YouTube, Instagram, TikTok, Facebook, X/Twitter, Spotify, Vimeo, SoundCloud and similar platforms. Their terms do not allow downloading their media, so these links are refused (and not charged).
• Private network addresses (localhost, 10.x, 192.168.x, ...) are refused.
• The default run timeout is 2 hours. Processing takes roughly 8 s per audio minute (Standard) or 22 s (Accurate) at the default 4 GB memory; doubling the memory roughly halves it. If a file cannot finish before the run's timeout, it is skipped up front (not charged) with a message telling you to raise the timeout or memory.
• No speaker labels (diarization) and no translation yet.

Your responsibility

Only submit audio you own or have permission to process, such as your own recordings, interviews with consent, or public podcast feeds for personal or research use. Transcripts of people's voices can contain personal data. You decide what you do with them and must handle them lawfully (e.g. GDPR).

Privacy

Files are downloaded into the run's temporary container, transcribed and deleted when the run ends. Transcripts are stored only in your run's dataset in your Apify account. The developer does not receive your audio or transcripts.

Open-source components

faster-whisper and CTranslate2 (MIT), Whisper base/small model weights (MIT, converted by SYSTRAN), FFmpeg (LGPL/GPL, used as an unmodified command-line tool), Apify SDK (Apache-2.0). This Actor is an independent product and is not affiliated with or endorsed by OpenAI.

Support

Open an issue on the Issues tab. Typical response time is 2–5 business days.