Speech & Audio to Text: Whisper Transcription with SRT/VTT

What does this speech-to-text Actor do?

It transcribes audio and video files from direct links (speech to text, plus subtitles) using an open-source Whisper speech-recognition model that runs inside the Actor. For each file you get:

• the full text;
• timestamped segments (start, end, text);
• ready-to-use SRT and VTT subtitles;
• the detected language, with a confidence value;
• the audio duration and the number of billed minutes.

There's no third-party speech API: audio is processed only inside the Actor run on Apify.

Typical uses: transcribing your podcast episodes from their RSS enclosure links, subtitling your own videos, turning meeting or interview recordings into searchable text, and preparing transcripts for LLM pipelines.

Why run it on Apify?

• Call it from your own code through the Apify API, or from AI agents through Apify's MCP server.
• Schedule runs, for example to transcribe each new episode, and connect results to webhooks and integrations.
• Download results as JSON, CSV or Excel.
• Batch many files in one run; a failed file doesn't stop the others.

How to use it

1. Add one or more media file URLs: direct links to mp3, m4a, wav, ogg, mp4, webm and similar files, e.g. from your own storage, signed cloud links or podcast feeds.
2. Choose the model:
   • base (default): fast and low-cost;
   • small: more accurate, priced higher.
3. Optionally set the language (e.g. en, de, fr). It's detected automatically otherwise.
4. Run the Actor and open the Output tab, or fetch the dataset through the API.

Pages on video platforms (YouTube, TikTok and similar) aren't supported. This Actor only downloads direct media files that you supply.

Input example

{ "audioUrls": ["https://example.com/episode-12.mp3"], "model": "base" }

Output

One dataset item per file:

• url (query string removed) and inputIndex;
• status, language, languageProbability;
• durationSec, billedMinutes;
• text;
• segments;
• srt and vtt (complete subtitle files as text).

Failed files come back as rows with status: "error" and a reason.

How much does it cost?

You pay per started minute of audio, plus a small start fee per run (charged per GB of memory) that covers loading the speech model. See the Pricing tab for current prices. The small model costs more per minute than base. Batching several files into one run spreads the start fee across them.

Apart from the start fee, you're never charged for:

• failed downloads;
• links that aren't media files;
• files longer than your Max minutes per file setting.

The Actor stops before any file that would exceed the maximum charge you set for the run.

Data handling

• Only public http(s) hosts are fetched. Private, loopback and internal addresses are refused, including after redirects.
• Each file is downloaded into the run's temporary storage, processed there, and deleted once that file is done.
• Results go to your run's Apify dataset and are kept according to your Apify account's data-retention settings.
• Query strings are removed from URLs in the output and the logs, so signed-URL tokens aren't copied. Use inputIndex to match each result to your input list.

Accuracy and limits

• Whisper base handles clear speech well, but makes occasional mistakes with names, place names and rare words. small is noticeably better. In our tests both still misheard some 19th-century place names in a poetry reading.
• Check important transcripts before publishing them.
• There's no speaker diarisation (it doesn't label who is speaking).
• Maximum 1 GB per file; the maximum minutes per file is configurable (default 180).
• Only transcribe recordings you have the right to process.