Audio & Video to Text — Transcription with SRT, VTT and RAG chunks

Transcribe your own audio and video files — meetings, interviews, classes, podcasts, WhatsApp voice notes, webinars — into text you can use right away:

• 📝 Clean text in paragraphs, plus every segment with its start and end time.
• 🎬 SRT and VTT subtitles ready to upload, with cues of at most two lines and about 7 seconds.
• 🧩 Ready-to-embed chunks for RAG that never cut a sentence segment and keep their start and end time, so an agent can cite "minute 12:40".
• 📄 Markdown and TXT files for each transcript, with download links in the result.
• 🇪🇸 Strong in Spanish and 90+ other languages with automatic detection. Add names and terms (brands, people, places) so they are spelled right.
• 🔒 Open-source Whisper running inside Apify: your audio is not sent to OpenAI or any third-party API.

How to use it

1. Paste direct URLs to your audio/video files, or upload files and select the storage they were uploaded to in Storage with uploaded files (the Actor runs with limited permissions and can only read the storage you choose).
2. Choose Accurate (recommended) or Fast draft, and the language (or leave auto-detect).
3. Run it. Leave the input empty to try a demo with a public-domain audio (not charged).

From code or an AI agent, call it through the Apify API or MCP with {"files": ["https://example.com/meeting.m4a"], "language": "es"}.

Formats: MP3, WAV, M4A, AAC, OGG, OPUS, FLAC, WEBM, MP4, MOV, MKV, AVI, WMA, AMR, 3GP. Up to 500 MB per file. Links to video platforms (YouTube, TikTok, Instagram…) are not supported: use direct links to files you own.

Input

Field	Default	What it does
files	—	Files or direct URLs.
storage	—	The key-value store with your uploaded files.
mode	accurate	accurate (Whisper small) or fast (Whisper base, cheaper, more mistakes).
language	auto	Spoken language, e.g. es, en, pt.
vocabulary	—	Names and terms that appear in the audio, comma separated.
output_mode	transcript	One item per transcript (file), per segments or per chunks.
chunk_tokens	400	Target chunk size (≈4 characters per token). 0 = no chunks.
chunk_overlap	40	Last segments repeated at the start of the next chunk.
skip_silence	true	Skips long silences and music (faster, avoids invented text).
save_files	true	Saves .srt, .vtt, .md and .txt per file.
max_minutes_per_file	240	Longer files are skipped and not charged.

Output (transcript mode)

{
  "file": "team-meeting.m4a",
  "status": "ok",
  "language": "es",
  "language_probability": 0.99,
  "duration": "42:17",
  "duration_s": 2537.4,
  "minutes": 43,
  "word_count": 6120,
  "text": "Buen día a todos, arrancamos con el presupuesto del trimestre…\n\nSegundo punto: logística…",
  "segments": [{"id": 1, "start": 0.0, "end": 4.2, "text": "Buen día a todos, arrancamos con el presupuesto del trimestre."}],
  "chunks": [{"chunk_id": 1, "start": 0.0, "end": 96.4, "start_ts": "0:00", "end_ts": "1:36", "tokens_est": 398, "text": "…"}],
  "files": {"srt": "https://api.apify.com/v2/key-value-stores/…/records/001-team-meeting.srt", "vtt": "…", "markdown": "…", "text": "…"},
  "mode": "accurate",
  "billing": {"audio-minute": 43},
  "warnings": []
}

In segments mode each item is one timestamped segment; in chunks mode each item is one chunk with start, end, start_ts, end_ts, tokens_est and text, ready for a vector database.

Pricing

Pay per audio minute (rounded up per file), only for files where speech was found:

Event	Price	When
audio-minute	$0.010	Each minute of audio transcribed in Accurate mode
audio-minute-fast	$0.004	Each minute of audio transcribed in Fast draft mode

Files without speech, failed downloads, unsupported files and the demo are not charged. A file that would go over the maximum cost per run you set is skipped before it is transcribed, so you never pay for a cut-off transcript.

Speed

Accurate mode processes about 3 minutes of audio per minute with the default 4 GB of memory; Fast mode about 5. Give the run more memory (8 or 16 GB) to go faster: the price per audio minute stays the same.

Limits (honest list)

• No speaker labels (who said what) yet.
• Very noisy audio, people talking over each other or heavy accents lower the accuracy, as with any speech model. Accurate mode handles them much better than Fast.
• Long silences are billed as part of the file's duration (they are skipped, not removed from the file).
• Only direct links to files: no YouTube, TikTok, Instagram or other platforms.

Privacy and permissions

• Network access only to download the files you give it. The Apify token is only ever sent to api.apify.com.
• Audio is processed in the run's container with open-source models (faster-whisper, MIT license; Whisper models, MIT license) and deleted when the run ends. Results stay in your own dataset and key-value store.
• Limited permissions: the Actor can only read the storage you select.

Feedback

Got a transcript with errors? Open an issue on the Actor page with a short public sample (or describe the audio) and it will be looked at.