Transcribe Audio & Video to Text: Speech to Text, SRT, VTT

Audio to text and video transcription in one Actor: speech to text for recordings, podcasts and videos you own, with subtitle files ready to use.

Give this Actor links to your own audio or video files, or a podcast RSS feed. It returns a clean transcript with timestamps, plus SRT and VTT subtitle files. It runs the open Whisper model on Apify, so you need no API key and no third-party account. You pay per minute of audio actually transcribed, and files that fail cost nothing. It does not download from YouTube, TikTok, Instagram, Spotify or other platforms; use your own files.

Quick start for AI agents and API users

• What it does: Transcribe audio or video to text with open Whisper: direct media links, uploads or podcast RSS episodes.
• When to use it: For speech-to-text or SRT/VTT subtitles.
• Example input (the form's prefill; it finishes in under a minute):

{
  "mediaUrls": [
    {
      "url": "https://upload.wikimedia.org/wikipedia/commons/5/50/Jfk_rice_university_we_choose_to_go_to_the_moon.ogg"
    }
  ],
  "model": "base",
  "language": "en",
  "maxMinutesPerFile": 1,
  "outputFormats": [
    "text",
    "srt",
    "vtt"
  ]
}

• Example output: one real dataset row from run xxzBALvRPT6T4waYD (2026-10-23), with long fields shortened and empty fields left out. Fields not shown are listed in the output table below.

{
  "source": "https://upload.wikimedia.org/wikipedia/commons/5/50/Jfk_rice_university_we_choose_to_go_to_the_moon.ogg",
  "url": "https://upload.wikimedia.org/wikipedia/commons/5/50/Jfk_rice_university_we_choose_to_go_to_the_moon.ogg",
  "inputType": "url",
  "status": "ok",
  "fileName": "Jfk_rice_university_we_choose_to_go_to_the_moon.ogg",
  "model": "base",
  "language": "en",
  "languageProbability": 1,
  "durationSecs": 1060.25,
  "transcribedSecs": 60.01,
  "minutesCharged": 1,
  "text": "Mr. Vice President, Governor, Congressman Thomas, Senator Wiley and Congressman Miller, Mr. Webb, Mr. Bell, scientist, distinguished gues...",
  "truncated": true,
  "srtUrl": "https://api.apify.com/v2/key-value-stores/aKcf0hTYL5ee6PzfN/records/transcript-0001.srt?signature=12ZAgDYRdtIlTUn8sozM6",
  "vttUrl": "https://api.apify.com/v2/key-value-stores/aKcf0hTYL5ee6PzfN/records/transcript-0001.vtt?signature=1gXqozGcBx12buiyOkTje"
}

• Cost (pay per event, Free-plan prices): File transcribed $0.005; Minute of audio transcribed $0.008. Failed items are not charged.
• From an AI agent: with the Apify MCP server, call call-actor with actor phaux/audio-video-transcriber, or load it as its own tool with https://mcp.apify.com?tools=phaux/audio-video-transcriber. Read results with get-dataset-items.

What does Audio & Video Transcriber do?

• Transcribes speech to text from audio and video files: mp3, m4a/AAC, wav, ogg/opus, flac, mp4, mov, webm, mkv and anything else FFmpeg can read.
• Three ways in: direct file links (your server, S3/GCS signed URLs, Dropbox direct links, Apify storage), file upload in the input form, or podcast RSS feeds (the newest episodes' audio enclosures).
• Outputs plain text, timestamped segments, optional word-level timestamps, and SRT / WebVTT subtitles (inline and as downloadable files in the key-value store).
• 99 languages with automatic language detection, or set the language for better accuracy.
• Reads only what it needs: with Max minutes per file it stops reading after the first N minutes, so long files stay cheap.
• Skips silence with voice-activity detection, which is faster and avoids made-up text in quiet parts.

Why use it?

• Predictable price: a small fee per file plus a price per minute transcribed. Failed, refused, silent or unreadable files are free.
• No API key, no extra account: the open-source Whisper model (faster-whisper, MIT) runs inside the Actor.
• Subtitles ready to use: SRT and VTT files you can upload to a video editor or player.
• Your data stays yours: the Actor keeps no transcript logs; results are only in your run's storage.

Use cases: meeting and interview recordings, lecture and webinar captions, podcast show notes and search, voice notes, call recordings you own, video subtitles, and feeding transcripts into an LLM or RAG pipeline.

What it does NOT do

It refuses links to YouTube, TikTok, Spotify, Apple Music / Apple Podcasts pages, SoundCloud, Instagram, Facebook, X/Twitter, Vimeo, Loom, Twitch and similar platforms (also when a link redirects there). Those platforms' terms forbid third-party downloading. Use a file you have the rights to. Web pages are refused too: the link must point to the media file itself.

How to use it

1. Paste direct links into Media file URLs, upload files under Upload files, or add Podcast RSS feeds.
2. Pick the Model: base (fast, default), small (more accurate on accents, noise and names) or tiny (rough drafts). The price is the same.
3. Set the Language (e.g. en) or leave auto.
4. For subtitles, keep srt / vtt in Output formats and optionally set Caption max characters per line (e.g. 42).
5. Optionally set Max minutes per file to cap cost, then run. Open the Transcripts or Subtitles (SRT / VTT) view, or download SRT/VTT from the links in each row.

Tutorial: make SRT and VTT captions for a video

This walk-through makes subtitle files for a real recording: the first minute of John F. Kennedy's 1962 Rice University speech (public domain, Wikimedia Commons). Everything below is copied from an actual run of this Actor (22 Oct 2026, small model, 54 s run time).

1. Input. Point the Actor at the file, ask for srt and vtt, and set a caption line length so each cue fits on screen:

{
  "mediaUrls": [{"url": "https://upload.wikimedia.org/wikipedia/commons/5/50/Jfk_rice_university_we_choose_to_go_to_the_moon.ogg"}],
  "model": "small",
  "language": "en",
  "maxMinutesPerFile": 1,
  "outputFormats": ["text", "srt", "vtt"],
  "captionMaxCharsPerLine": 42,
  "captionMaxLines": 2
}

2. Run it, then open the Subtitles (SRT / VTT) view of the dataset. Each row has srtUrl and vttUrl download links (files transcript-0001.srt and transcript-0001.vtt in the run's key-value store) and the subtitle text inline in srt / vtt. This minute became 15 cues.

3. The SRT file (first 6 cues as returned):

1
00:00:02,450 --> 00:00:06,290
President Pipps,
Mr. Vice President, Governor,

2
00:00:08,150 --> 00:00:11,450
Congressman Thomas,
Senator Wiley and Congressman

3
00:00:11,450 --> 00:00:16,430
Miller, Mr. Webb, Mr. Bell, scientists,

4
00:00:17,330 --> 00:00:18,830
distinguished guests,
ladies and gentlemen.

5
00:00:19,530 --> 00:00:25,810
I appreciate your President having made me
an honorary visiting professor and I will

6
00:00:25,810 --> 00:00:29,050
assure you that my first
lecture will be very brief.

(The speaker actually says "President Pitzer". Speech models often mishear names. Check names before you publish captions.)

4. The same cues as WebVTT, for HTML5 players and most web video tools:

WEBVTT

00:00:02.450 --> 00:00:06.290
President Pipps,
Mr. Vice President, Governor,

00:00:08.150 --> 00:00:11.450
Congressman Thomas,
Senator Wiley and Congressman

5. Use the files.

• YouTube Studio: Subtitles → Add language → Upload file → "With timing" → pick the .srt.
• Premiere Pro / DaVinci Resolve / Final Cut (via SRT import): import the .srt as a captions track.
• HTML5 video: <track kind="captions" src="transcript-0001.vtt" srclang="en" label="English" default>.
• Burn in with FFmpeg: ffmpeg -i video.mp4 -vf subtitles=transcript-0001.srt out.mp4.

Caption length: with and without captionMaxCharsPerLine. If you leave it at 0 (the default), you get one cue per Whisper segment, as before. Those cues are fine for search and notes, but they are often too long to read on screen. Here is what the same speech gave on the default setting (prefill run, base model):

2
00:00:11,750 --> 00:00:19,910
Miller, Mr. Webb, Mr. Bell, scientists, distinguished guests and ladies and gentlemen.

With captionMaxCharsPerLine: 42 the Actor turns on word timestamps and re-cuts every segment at word boundaries. Each line is at most 42 characters, there are at most captionMaxLines lines per cue, and each cue lasts about 7 seconds or less. It prefers to break after punctuation and never splits "Mr. Bell". Good values are 42 (the common TV/streaming guideline), 32 (phones and vertical video) and captionMaxLines: 1 (single-line social clips). The cues are what change. text and segments keep the same format, though word timestamps can shift segment boundaries a little. The price is the same.

Input example

{
  "mediaUrls": [{"url": "https://upload.wikimedia.org/wikipedia/commons/5/50/Jfk_rice_university_we_choose_to_go_to_the_moon.ogg"}],
  "model": "base",
  "language": "en",
  "maxMinutesPerFile": 1,
  "outputFormats": ["text", "segments", "srt", "vtt"]
}

Empty input runs a small example: the first minute of John F. Kennedy's 1962 Rice University "We choose to go to the Moon" speech (public domain, Wikimedia Commons).

Field	Default	Description
mediaUrls	—	Direct links to audio/video files.
uploadedFiles	—	Files uploaded through the form (stored in your key-value store).
podcastRssUrls	—	Podcast RSS/Atom feeds; newest episodes first.
maxEpisodesPerFeed	1	Episodes per feed (1–100).
model	base	tiny, base or small.
language	auto	2-letter language code or auto.
maxMinutesPerFile	120	Transcribe and charge only the first N minutes (1–300).
outputFormats	text, segments, srt, vtt	Any of text, segments, words, srt, vtt.
captionMaxCharsPerLine	0 (off)	Re-cut SRT/VTT cues so each line has at most N characters (e.g. 42 or 32). 0 = one cue per segment, as before.
captionMaxLines	2	Lines per cue when the line limit is on (1–3).
skipSilence	true	Voice-activity detection.
maxFiles	0	Stop after N files (0 = all).
timeBudgetSecs	0	Stop after N seconds; a file cut short is charged only for what was transcribed.

Output example

{
  "source": "https://upload.wikimedia.org/wikipedia/commons/5/50/Jfk_rice_university_we_choose_to_go_to_the_moon.ogg",
  "status": "ok",
  "fileName": "Jfk_rice_university_we_choose_to_go_to_the_moon.ogg",
  "model": "base",
  "language": "en",
  "durationSecs": 1060.25,
  "transcribedSecs": 60.01,
  "truncated": true,
  "minutesCharged": 1,
  "text": "... Mr. Vice President, Governor, Congressman Thomas, Senator Wiley and Congressman Miller, Mr. Webb, Mr. Bell, scientists, distinguished guests and ladies and gentlemen ...",
  "segments": [{"start": 2.45, "end": 11.73, "text": "..."}],
  "srt": "1\n00:00:02,450 --> 00:00:11,730\n...",
  "srtUrl": "https://api.apify.com/v2/key-value-stores/.../records/transcript-0001.srt",
  "vttUrl": "https://api.apify.com/v2/key-value-stores/.../records/transcript-0001.vtt",
  "error": null
}

Failed or refused files get "status": "failed" and an error such as {"code": "platform_not_supported", ...}, not_media, not_found, access_denied, robots_disallowed or unreadable_media. They are not charged. RUN_STATS and ERRORS records are in the key-value store.

Pricing

Pay per event:

Event	Price
File transcribed (once per file with a transcript)	$0.005
Minute of audio transcribed (per started minute)	$0.008

Example: a 30-minute recording costs $0.005 + 30 × $0.008 = $0.245. A 10-file batch of 3-minute voice notes costs 10 × ($0.005 + 3 × $0.008) = $0.29. Minutes are counted from the audio actually transcribed (capped by Max minutes per file), rounded up to whole minutes per file. Set a maximum total charge on the run and the Actor stops before it.

Tips and limits

• small is about 3× slower than base; for long files give the run more time (and 4 GB memory is faster).
• Default memory is 2 GB. For files longer than ~2 hours use 4 GB.
• Podcast feeds and enclosure hosts are checked against robots.txt; some ad-tracking redirect hosts disallow bots, and those episodes are reported as robots_disallowed (free).
• Private links (login, 401/403) can't be read; use a public or signed link, or upload the file.

FAQ

Can it transcribe YouTube, TikTok or Spotify links? No. Platform links are refused because those platforms' terms forbid third-party downloading. Use a direct link to a file you have the rights to, or upload the file.

Which languages does it support? 99 languages. Leave the language on auto to detect it, or set it (for example en) for better accuracy.

Do I need an OpenAI or other API key? No. The open Whisper model runs inside the Actor.

How much does a one-hour recording cost? $0.005 per file plus $0.008 per minute, so 60 minutes cost $0.005 + 60 × $0.008 = $0.485. Use Max minutes per file to cap it.

Which model should I pick? base is the fast default. small is more accurate on accents, noise and names but about 3× slower. tiny is for rough drafts. The price is the same.

How do I get subtitles? Include srt and/or vtt in Output formats. Each row has the subtitle text plus srtUrl and vttUrl download links. For on-screen captions, also set Caption max characters per line (for example 42). See the tutorial above.

Am I charged for files that fail? No. Failed, refused, silent or unreadable files are free.

Something isn't working? Open an issue on the Actor's Issues tab with the run ID.

Legal

Transcribe only content you own or have the right to process. Copyright in the audio and the transcript stays with its owner. The Actor stores nothing beyond your run's own storage. FFmpeg (Debian build) is used to decode media; Whisper weights by OpenAI / Systran conversions are MIT-licensed.