Audio & Video Transcriber: Word Timestamps, SRT, VTT

Paste audio or video links and get the transcript with word-level timestamps: each word carries its start time, end time and a confidence score. You also get SRT and VTT subtitles and the plain text. Send one link or a few hundred. Files can run up to 3 hours, three process at a time, and a 44-minute interview takes under a minute.

What you get for each file

Field	What it holds
text	The full transcript
words	Each word with s (start, seconds), e (end) and p (confidence, 0 to 1)
segments	Sentence-level chunks with start and end times
srt, vtt	Subtitle files, split at natural pauses
language	The language code it detected, or the one you set
duration_s, word_count	Audio length and number of words
flags	Each change the quality guards made, with timestamps (see below)

Guards against Whisper's known failure modes

Whisper sometimes invents phrases over silence ("Thanks for watching!", "Subtitles by...") or repeats one line over and over. This Actor drops an invented phrase when the model's own confidence in it is low, and collapses repetition loops. You see each fix in flags, with the original text and its timestamps, so you can review or undo it.

How to use it

1. Paste your links into Audio or video URLs. Direct file links work best (mp3, wav, m4a, ogg, flac, mp4, mov, webm and more).
2. Set the language if you know it. Leave it empty and the Actor detects it per file.
3. Click Start. The Transcripts tab in Output shows one row per file, and the Subtitles tab holds the SRT and VTT files.

You can also call it through the Apify API, run it on a schedule, or connect it to Make, Zapier, n8n and other Apify integrations.

How much does transcription cost?

$0.008 per audio minute, rounded up per file. Failed downloads and unreadable files cost nothing.

Audio	Cost
30-second clip	$0.008
44-minute interview	$0.35
1-hour podcast episode	$0.48
100 one-hour episodes	$48.00

Example

Input, using the 30-second sample prefilled in the form (a public-domain NASA interview with astronaut Bob Behnken):

{ "urls": [{ "url": "https://patchwell.io/samples/nasa_interview_30s.mp3" }] }

Real output from that file, shortened:

{
  "url": "https://patchwell.io/samples/nasa_interview_30s.mp3",
  "ok": true,
  "language": "en",
  "duration_s": 32.08,
  "word_count": 90,
  "text": "Sure. My name is Bob Behnken. And to be clear, you go by Bob, not Robert, right? I do. Okay, great...",
  "words": [
    { "w": "Sure.", "s": 0.0, "e": 0.3, "p": 0.782 },
    { "w": "My", "s": 0.46, "e": 0.52, "p": 0.998 },
    { "w": "name", "s": 0.52, "e": 0.66, "p": 1.0 }
  ],
  "srt": "1\n00:00:00,000 --> 00:00:00,300\nSure.\n\n2\n00:00:00,460 --> 00:00:01,740\nMy name is Bob Behnken.\n...",
  "flags": []
}

Speed

On the full 44-minute version of that interview (2,626 seconds of audio), the run took 49 seconds end to end and returned 5,613 words. After a quiet period, the first run needs 10 to 30 seconds more while a GPU starts.

Use it from an AI agent

Add this Actor to Claude, Cursor or any MCP client through Apify's MCP server. With word timings, your agent can quote a recording to the second: "at 12:04 the guest said..."

FAQ

Which model does it use?

Whisper large-v3-turbo (through faster-whisper) on NVIDIA GPUs, plus the quality guards above.

Can it transcribe YouTube links?

No. It needs a direct link to a media file. Run a YouTube downloader Actor first and pass its file URL here, or use files you host.

Does it identify speakers?

Not yet. Speaker labels are on the roadmap. Today you get words, timings and confidence.

Which languages work?

The 99 languages Whisper supports. English and other widely spoken languages come out most accurate. For anything critical, check the p confidence values.

Is my audio stored?

The Actor downloads each file into a temporary container, transcribes it, and discards both. Patchwell keeps no copies. Results stay in your own Apify storage.

Other Patchwell tools

• AI Web Research with Cited Sources: answers where each sentence cites its source page.
• CSV & Excel Data Cleaner with Change Log: clean messy spreadsheets or any Actor's dataset, with each fix logged.

Questions or a custom need: patchwell.support@gmail.com