Audio & Video Transcriber: Speech to Text, SRT & VTT Subtitles (Whisper)

Turn audio and video files into text with OpenAI's open-source Whisper speech recognition model. Paste links to MP3, MP4, WAV, M4A, WebM, OGG, FLAC or MOV files (or upload one), and get a clean transcript, timestamped segments, SRT and WebVTT subtitle files, optional word-level timestamps and translation to English. 99 languages with automatic language detection. No API key, no OpenAI account, no setup. You pay per minute of audio, from $0.005 per minute.

What you can use it for

• Podcast and interview transcripts: searchable text and show notes from episodes and recorded calls.
• Subtitles and captions: SRT/VTT files for videos, courses, webinars and social clips, ready to upload to any video player or editor.
• Meeting and lecture notes: transcribe recordings, then summarize them with ChatGPT, Claude or your own LLM.
• AI and RAG pipelines: feed transcripts with timestamps into vector databases, agents and n8n/Make/Zapier workflows.
• Translate speech to English: Spanish, Hebrew, Arabic, French, German, Hindi and 90+ other languages to English text in one step.
• Content repurposing: turn videos into blog posts, quotes and newsletter material.

How to use it

1. Paste one or more direct file links (or upload a file). Google Drive files shared as Anyone with the link and Dropbox links are converted automatically.
2. Pick Fast or Accurate quality and, optionally, the spoken language.
3. Click Start. Each file becomes one result row with the full text, plus links to the SRT, VTT and TXT files.

Fast or Accurate?

	Fast	Accurate
Model	Whisper base (int8)	Whisper small (int8, beam search 5)
Best for	Clear speech, English and major languages	Accents, noisy audio, names, non-English speech
Price	$0.005 / audio minute	$0.010 / audio minute

Pricing example

Billing is per started minute of audio in each file (a 30-minute podcast = 30 minutes), plus a tiny start fee. Files that fail to download or contain no speech are not charged.

Job	Fast	Accurate
One 10-minute interview	$0.05	$0.10
Ten 30-minute podcast episodes (300 min)	$1.50	$3.00
100 hours of lectures (6,000 min)	$30	$60

Features

• Every common format: anything FFmpeg can read, including video files (the audio track is extracted automatically).
• 99 languages with auto-detect, or force a language for better accuracy.
• Translate to English (Whisper's built-in translation task).
• SRT, WebVTT, TXT and JSON files saved to the run's key-value store, with direct links in each result.
• Word-level timestamps with confidence scores (for karaoke-style captions, editing tools and search).
• Silence skipping (VAD): detects speech first, which avoids the "phantom text" Whisper can produce on silence or music.
• Vocabulary hint: give it names, brands or jargon to improve spelling.
• Budget safe: before transcribing, it checks your run's max cost and skips files that would exceed it.
• Private by design: files are processed inside your run and not stored anywhere else.

Input example

{
  "fileUrls": [{"url": "https://www.archive.org/download/count_monte_cristo_0711_librivox/count_of_monte_cristo_001_dumas.mp3"}],
  "quality": "fast",
  "language": "auto",
  "task": "transcribe",
  "outputFormats": ["srt", "vtt", "txt"],
  "wordTimestamps": false,
  "maxMinutesPerFile": 180
}

Output example (one item per file)

{
  "url": "https://www.archive.org/download/count_monte_cristo_0711_librivox/count_of_monte_cristo_001_dumas.mp3",
  "fileSizeMB": 18,
  "durationSeconds": 1179.5,
  "language": "en",
  "languageProbability": 0.962,
  "quality": "fast",
  "model": "whisper-base",
  "task": "transcribe",
  "transcribedSeconds": 300,
  "truncated": true,
  "billedMinutes": 5,
  "processingSeconds": 34.9,
  "segmentCount": 62,
  "text": "This is a Librevox recording. All Librevox recordings are in the public domain and for more information or to volunteer visit Librevox.org\n\nRecording by Christian Luoma, greenkri.com of the Count of Monte Cristo by Alexander Dumas ...",
  "wordCount": 728,
  "srtUrl": "https://api.apify.com/v2/key-value-stores/<store-id>/records/001-count_of_monte_cristo_001_dumas.srt",
  "vttUrl": "https://api.apify.com/v2/key-value-stores/<store-id>/records/001-count_of_monte_cristo_001_dumas.vtt",
  "txtUrl": "https://api.apify.com/v2/key-value-stores/<store-id>/records/001-count_of_monte_cristo_001_dumas.txt",
  "segments": [
    {
      "start": 0.66,
      "end": 10.86,
      "text": "This is a Librevox recording. All Librevox recordings are in the public domain and for more information or to volunteer visit Librevox.org"
    },
    {
      "start": 12.86,
      "end": 22.3,
      "text": "Recording by Christian Luoma, greenkri.com of the Count of Monte Cristo by Alexander Dumas"
    }
  ]
}

This example was run with Max minutes per file = 5 on a 19.7-minute file, so only 5 minutes were transcribed and billed (truncated: true). With Word-level timestamps on, every segment also has a words array with start, end, word and probability.

Failed files produce a row with url and error (for example File has no audio track or The link returned a web page, not a media file) and are not charged.

Use it from code, n8n or AI agents

Call it through the Apify API (POST https://api.apify.com/v2/acts/manageable_gully~audio-video-transcriber/run-sync-get-dataset-items), the Apify Python/JavaScript clients, the Apify n8n/Make/Zapier integrations, or as a tool for AI agents through the Apify MCP server.

FAQ

Which links work? Direct links to media files (they should download the file, not open a web page), Apify key-value store records, S3/GCS signed URLs, Google Drive (shared publicly) and Dropbox share links.

Can I transcribe YouTube, TikTok or Instagram links? No. Those are web pages, and downloading from them breaks their terms. Download your own video file and pass its link instead.

How long can a file be? Up to 240 minutes and 1 GB per file. Use Max minutes per file to cap cost on long recordings.

How accurate is it? It's Whisper, the same open-source model family used by many paid services. Accuracy depends on audio quality; use Accurate for difficult audio and set the language when you know it.

Speaker labels (diarization)? Not yet. Segments are timestamped but not split by speaker.

Is my audio used for training? No. Transcription runs locally inside the Actor; nothing is sent to third-party AI APIs.

More tools from WebIntel Tools

• PDF & Image OCR to Text/Markdown
• Web Page to Markdown for LLMs
• URL to PDF & Screenshot
• Website Health Audit