Speech to Text: Audio, Video & Podcast Transcriber (Whisper)

Transcribe audio files, video files, podcast RSS feeds and Apple Podcasts episodes to text in bulk with OpenAI's open-source Whisper model, running inside the Actor. No OpenAI key, no outside API, and your audio is never sent to a third-party service.

• Any media link: MP3, M4A, AAC, WAV, FLAC, OGG, OPUS, WMA, AMR, MP4, MOV, MKV, WEBM, AVI, 3GP and HLS streams (.m3u8). Video files are fine; only the audio track is used
• Podcasts built in: paste a podcast RSS feed or an Apple Podcasts show or episode link and the newest episodes are found and transcribed, with episode title, show name and publish date
• 99 languages with automatic language detection per file, or set the language yourself
• Translate to English from any supported language in the same pass
• Full transcript plus timestamped segments (start, end, confidence), optional word-level timestamps, and ready-made SRT and WebVTT subtitles
• Three Whisper models: Tiny (fastest), Base (default, good for podcasts and talks) and Small (best for accents, noisy audio and non-English speech)
• Vocabulary hint to spell names, brands and jargon correctly
• Silence and music skipping (voice activity detection) for faster runs and fewer hallucinated lines
• Long files (multi-hour podcasts, lectures, meetings) are streamed and transcribed in windows split at natural pauses
• Silent files and failed downloads are free, and each failure is listed with its reason
• Works with the Apple Podcasts Scraper and RSS Feed Reader: pass their dataset ID to transcribe every episode they found

Pricing: from $0.002 per audio minute

	This Actor	Other Whisper Actors on Apify
Tiny model	$0.002 / min ($0.12 per hour)	–
Base model (default)	$0.004 / min ($0.24 per hour)	$0.007 to $0.05 / min
Small model	$0.012 / min ($0.72 per hour)	$0.016 / min and up
Per-file or start fees	none (only $0.0005 per GB at run start)	up to $0.005 per file
Silent files, failed URLs	free	often charged

You pay per started minute of audio actually transcribed (with a 1-second grace). A 45-minute podcast episode costs $0.18 with Base; 100 hours of audio costs $24. Use Max minutes per file for cheap previews and set a maximum cost per run in Apify: the run stops cleanly at your limit, trimming the last file if needed.

Use cases

• Podcast transcripts for show notes, SEO pages, blog posts, newsletters and quotes
• Subtitles and captions (SRT / VTT) for videos, courses and social clips
• AI and RAG pipelines: turn audio libraries into searchable text for ChatGPT, Claude, LangChain or a vector database
• Media monitoring and research: track what podcasts and broadcasters say about a brand, topic or competitor
• Meetings, interviews, lectures and webinars stored as files or on a public link
• Translation: English text from Spanish, German, French, Arabic, Hindi, Japanese and 90+ other languages

Input

Field	What it does
urls	Media links, podcast RSS feeds or Apple Podcasts links. Empty input transcribes an 11-second sample
maxEpisodesPerFeed	Newest episodes per feed or show (default 3)
model	base (default), tiny or small
language	ISO code (en, de, es, ar...); empty = auto-detect
translateToEnglish	English translation instead of the original language
maxMinutesPerFile	Only the first N minutes of each file (0 = all)
vocabulary	Names and terms to spell correctly
includeSegments, includeSrt, includeVtt, includeWords	Choose the output detail
skipSilence	Voice activity detection (on by default)
accurate	Beam search for slightly better accuracy (about 2x slower, same price)
datasetId, urlField	Read media URLs from a dataset (audioUrl, enclosureUrl, mediaUrl, episodeUrl, url are detected)

{
  "urls": [
    "https://feeds.npr.org/510289/podcast.xml",
    "https://podcasts.apple.com/us/podcast/the-daily/id1200361736"
  ],
  "maxEpisodesPerFeed": 2,
  "model": "base",
  "includeSrt": true
}

Output

One row per media file:

{
  "status": "ok",
  "title": "The history of light (classic)",
  "podcast": "Planet Money",
  "published": "Wed, 04 Nov 2026 08:00:00 +0000",
  "mediaUrl": "https://.../episode.mp3",
  "feedUrl": "https://feeds.npr.org/510289/podcast.xml",
  "model": "base",
  "language": "en",
  "languageProbability": 0.999,
  "durationSec": 1503.2,
  "speechSec": 1471.9,
  "wordCount": 4120,
  "chargedMinutes": 26,
  "processingSec": 131.4,
  "text": "This message comes from ... ",
  "segments": [{ "start": 0.0, "end": 6.64, "text": "This message comes from ...", "avgLogprob": -0.21, "noSpeechProb": 0.01 }],
  "srt": "1\n00:00:00,000 --> 00:00:06,640\nThis message comes from ...\n"
}

Rows with status error (unreachable link, no audio track) or no-speech are free. A SUMMARY record in the key-value store has file, minute and error counts.

Speed and memory

Whisper runs on CPU. On the default 4 GB (one CPU core), Base transcribes about 11 minutes of audio per minute, Tiny about twice as fast, Small about 3.5x real time. More memory adds CPU cores and speeds up each file; the price per audio minute stays the same.

Supported sources and limits

• Public, directly downloadable media files and HLS streams, public podcast RSS feeds, and Apple Podcasts shows and episodes (via their public RSS feed)
• Not supported: DRM-protected audio (Spotify, Audible), pages that need a login, and video platforms such as YouTube or TikTok. For YouTube use the YouTube Transcript Scraper
• Speaker labels (diarization) are not included
• Only transcribe media you have the right to process

FAQ

Is this the same Whisper as the OpenAI API? It is the same open-source Whisper model family (tiny, base, small), run locally with the CTranslate2 engine in 8-bit mode. Results match the original models closely, and no data leaves the Apify platform.

How accurate is it? Base is reliable for clear speech such as podcasts, audiobooks and talks. For heavy accents, background noise, phone audio or non-English speech choose Small, and add a vocabulary hint for names.

Can I transcribe a whole podcast archive? Yes: set maxEpisodesPerFeed up to 1,000, or scrape episode URLs with the Apple Podcasts Scraper and pass its dataset ID. Files are downloaded and decoded in parallel with transcription.

What if a link fails? The row is pushed with status: "error" and the reason, and you are not charged.