Audio & Video Transcriber (Whisper speech to text, SRT, VTT)

Turn direct audio and video file URLs into text with timestamps. The Actor runs the open-source Whisper model (through faster-whisper) inside the container, so there is no third-party API key and your files are not sent to any other AI service.

• Works with MP3, M4A, AAC, WAV, FLAC, OGG, OPUS, MP4, MOV, WebM, MKV and most other formats FFmpeg can read.
• About 99 languages with automatic language detection per file, or set the language yourself.
• Translate to English from any language with one switch.
• One dataset row per file: full text, language, length, word count and timestamped segments.
• Optional SRT, WebVTT and TXT subtitle files with a link in each row.
• Only new files mode for scheduled runs over a list of podcast or meeting recording URLs.
• Files that fail are never charged: broken links, web pages instead of files, files without audio, files with no speech, or files above your size and length limits.

Direct file URLs only. The URL must return the media file itself (for example https://example.org/episode-12.mp3). Pages of YouTube, TikTok, Instagram, Facebook, X, Vimeo, Spotify and similar platforms are rejected, because their terms do not allow downloading. Use podcast RSS enclosure links, your own cloud storage links, or files you uploaded to an Apify key-value store (pick the store and list the keys).

How to use

1. Paste one or more direct file links into Media file URLs, or pick an Apify key-value store that holds your files and list the record keys.
2. Pick a Whisper model: tiny (fastest), base (balanced, default) or small (most accurate).
3. Optionally tick SRT, WebVTT or Plain text under Subtitle files.
4. Run. Download the transcripts as JSON, CSV or Excel from the dataset, and the subtitle files from the key-value store.

If you start the Actor with an empty input, it transcribes a 24-second public domain sample (Neil Armstrong on the Moon, NASA, via Wikimedia Commons) so you can see the output format.

Input example

{
  "mediaUrls": [
    "https://upload.wikimedia.org/wikipedia/commons/3/30/LibriVox_-_Everrett_Copy_of_the_Gettysburg_Address_-_Michael_Scherer.ogg"
  ],
  "model": "base",
  "subtitleFormats": ["srt"]
}

Output example

Real output for the input above (shortened):

{
  "url": "https://upload.wikimedia.org/wikipedia/commons/3/30/LibriVox_-_Everrett_Copy_of_the_Gettysburg_Address_-_Michael_Scherer.ogg",
  "fileName": "LibriVox_-_Everrett_Copy_of_the_Gettysburg_Address_-_Michael_Scherer.ogg",
  "language": "en",
  "languageProbability": 0.993,
  "task": "transcribe",
  "model": "base",
  "durationSeconds": 133.23,
  "billedMinutes": 3,
  "text": "This is a Libravox Recording. All Libravox recordings are in the public domain. ... Four score and seven years ago, our fathers brought forth upon this continent a new nation, conceived in liberty, ...",
  "wordCount": 302,
  "segmentCount": 27,
  "segments": [
    { "start": 1.62, "end": 6.62, "text": "This is a Libravox Recording. All Libravox recordings are in the public domain." },
    { "start": 6.62, "end": 11.62, "text": "For more information or to volunteer, please visit Libravox.org." }
  ],
  "srtUrl": "https://api.apify.com/v2/key-value-stores/<id>/records/001-LibriVox_-_Everrett_Copy_of_the_Gettysburg_Address_-_Michael_Scherer.srt",
  "processingSeconds": 9.4,
  "transcribedAt": "2026-10-27T04:22:44+00:00"
}

Notice "Libravox" instead of "LibriVox": Whisper writes what it hears. The small model got the name right in the same test.

The SUMMARY record in the key-value store lists every skipped file with the reason.

Pricing

Pay per event: $0.0015 per audio minute transcribed ($1.50 per 1,000 minutes), rounded up per file (a 40-second clip is 1 minute, a 12 minute 4 second file is 13 minutes). Store discounts: Bronze $0.00135, Silver $0.0012, Gold and above $0.00105 per minute. Skipped and failed files are free.

Platform usage (compute) is billed to your Apify account as for any Actor. Measured with the default 4 GB memory:

Model	Speed (audio minutes per run minute)	Compute per audio minute
tiny	about 20	about 0.003 CU
base (default)	about 7	about 0.009 CU
small	about 3	about 0.02 CU

Each run also has about 10 seconds of start-up (loading the model). Example from our test: a 12 minute 4 second MP3 with base took 107 seconds and 0.12 CU, so the total is $0.02 for 13 minutes plus 0.12 CU of compute (about $0.04 to $0.05 at $0.3 to $0.4 per CU), about $0.005 per audio minute all in. With tiny it is cheaper, with small about twice as much.

Tips

• Clear speech in a common language: base is usually enough. Accents, noise or less common languages: try small.
• Set Language if you already know it. This avoids rare wrong guesses on very short or music-heavy files.
• Skip silence is on by default. It speeds up long recordings with pauses and reduces made-up text in silent parts. Billing always uses the full file length.
• Use Only new files with a schedule to transcribe only new episodes from a list of URLs. The state is kept in a named key-value store (Monitor name).

FAQ

Can I transcribe a YouTube or TikTok link? No. This Actor only accepts direct file URLs. Platform pages are rejected immediately and not charged.

Does it identify speakers? No, speaker labels (diarization) are not included.

How long can a file be? Up to 600 minutes per file and 2 GB, with defaults of 240 minutes and 500 MB that you can raise.

Is my audio sent to OpenAI or another API? No. The model runs inside the Actor container. Files are deleted from the container after each transcription.

Can I call it from code or an AI agent? Yes, like any Apify Actor: the Apify API, the JavaScript and Python clients, Make, Zapier, n8n or the Apify MCP server.

Licenses

Whisper model weights and code: MIT (OpenAI). faster-whisper and the CTranslate2 model conversions: MIT (SYSTRAN). Sample audio: public domain (NASA). You are responsible for having the right to transcribe the files you submit.

Made by Ctrio Lab. Questions or feature requests: open an issue on the Actor page or write to help@ctriolab.com.