Audio & Video Transcription (Whisper) for Text and Subtitles avatar

Audio & Video Transcription (Whisper) for Text and Subtitles

Pricing

from $4.80 / 1,000 audio minute (fast)s

Go to Apify Store
Audio & Video Transcription (Whisper) for Text and Subtitles

Audio & Video Transcription (Whisper) for Text and Subtitles

Audio and video transcription with Whisper: turn MP3, WAV, M4A, MP4 and MOV files, or another Actor's dataset, into text with timestamps and SRT and VTT subtitles. About 100 languages, with a Whisper large-v3-turbo option. No API key or login. $6 per 1,000 audio minutes. Maintained by CleanScrape.

Pricing

from $4.80 / 1,000 audio minute (fast)s

Rating

0.0

(0)

Developer

CleanScrape

CleanScrape

Maintained by Community

Actor stats

0

Bookmarked

1

Total users

0

Monthly active users

an hour ago

Last modified

Features

Limited permissions

Share

Turn audio and video files into text. Paste direct links to MP3, M4A, WAV, FLAC, OGG, MP4, MOV, WebM or MKV files and get the full transcript with timestamps, plus TXT, SRT and VTT files for each one. It uses Whisper speech to text, works in about 100 languages and needs no API key or login.

On Fast and Accurate, Whisper runs inside the Actor on Apify: your audio is downloaded, transcribed and deleted there and is not sent to any other service. Best uses the larger Whisper large-v3-turbo model on Cloudflare Workers AI, at the price of Fast.

Price: $6 per 1,000 audio minutes on Fast and Best, $15 per 1,000 on Accurate. A one-hour recording is $0.36 or $0.90. No start fee. Files that fail are not charged.

Watch the walkthrough

NASA's recruiting film becomes text and SRT subtitles in 15 seconds, and an 8.5-minute audiobook chapter is transcribed in 28 seconds with Whisper large-v3-turbo. This 83-second demo uses real runs: paste a link, compare Fast and Best, and send another Actor's results straight into it.

Start with a small run

  1. The form opens with a short sample video. Keep it, or paste your own links under Audio and video links, one per line.
  2. Choose Fast, Best or Accurate.
  3. Click Start. Open Transcripts in the Output tab for the text, and Transcript files for the TXT, SRT and VTT files.

The sample is 67 seconds long. Length is rounded up to full minutes, so it costs 2 minutes: $0.012.

Choose a ready-to-run example

Each example opens with its input filled in. Change the links and run it.

ExampleWhat it does
Turn a video into text and SRT subtitlesA short MP4 video to text with timestamps, plus TXT, SRT and VTT files. About $0.01.
Transcribe audio with Whisper large-v3-turboA 9-minute MP3 on Best quality, done in under a minute. About $0.05.
Transcribe several audio files in one runMP3, M4A, WAV and FLAC files in one run, one row per file. About $0.05.

Which quality to choose

FastBestAccurate
ModelWhisper baseWhisper large-v3-turboWhisper small
Where the audio goesStays on ApifyCloudflare Workers AI, in short piecesStays on Apify
Good forClear speech, meetings, lectures, voice notesMost recordings, including accents, noise and other languagesAccents, noise and names when the audio must stay on Apify
Time for one hour of audioabout 10 to 15 minutesabout 2 to 3 minutesabout 30 minutes
Price$0.006 per audio minute$0.006 per audio minute$0.015 per audio minute

Best has a daily capacity. When it is used up or Cloudflare is unavailable, files run as Fast on Apify instead, at the same price, and model shows base, or large-v3-turbo and base for a file that used both. On Apify's Free plan, Best covers the first 30 minutes of audio in a run and the rest runs as Fast.

Giving the Language skips detection and avoids a wrong guess on short or noisy files. Names and terms to listen for helps Whisper spell product names, people and technical words.

  • A direct link to the file, such as https://example.com/talk.mp4. Links without a file ending work too when the server sends an audio or video file.
  • Dropbox and Google Drive share links, when the file is shared with anyone who has the link.
  • Signed links from cloud storage. Their query part is left out of anything the run shows.

Not supported: YouTube, TikTok, Instagram, Vimeo and Spotify page links (these are pages, not files), files behind a login, and podcast feeds. For podcast RSS feeds and Apple Podcasts links use Audio Transcription and Speech to Text for Podcasts.

Transcribe what another Actor found

Under Links from another Actor, choose a dataset: the results of a TikTok, Instagram, YouTube-style or podcast scraper, or any list of file links. The Actor finds the field that holds the file link (audioUrl, videoUrl, mediaUrl, downloadUrl and similar names first, then url), or reads the field you name, such as media.downloadUrl. Up to 200 files per run; the run message says if the dataset had more. Remove the sample link from Audio and video links when you use a dataset.

To transcribe every new result automatically, add this Actor as an integration on the other Actor or task: open its Integrations tab, choose Audio & Video Transcription (Whisper), and keep the prefilled input {"datasetId": "{{resource.defaultDatasetId}}"}. Each finished run of the other Actor then starts a transcription of its results.

Links must lead to the file itself. Page links (a TikTok or YouTube page) are refused one by one with the reason, and are not charged.

Use it from the API

from apify_client import ApifyClient
client = ApifyClient("<YOUR_API_TOKEN>")
run = client.actor("cleanscrape/audio-video-transcription-whisper").call(run_input={
"files": ["https://example.com/meeting.mp4", "https://example.com/interview.mp3"],
"quality": "fast",
"language": "English",
})
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
print(row["fileName"], row["duration"], row["wordCount"])
print(row["srtUrl"])

Input

FieldDefaultWhat it does
filesRequired unless datasetId is setDirect links to audio or video files, up to 200 per run
qualityfastfast (Whisper base), best (Whisper large-v3-turbo on Cloudflare) or accurate (Whisper small)
languageDetectedSpoken language as a name or code, such as English, fi, Japanese
vocabularyNoneNames and terms to listen for, separated by commas
datasetIdNoneA dataset with file links, such as another Actor's results
datasetFieldFoundThe dataset field that holds the link, such as videoUrl
maxMinutesPerFile240Longer files are skipped and not charged
maxFileMb1024Larger files are skipped and not charged

Output

One row per file.

FieldWhat it is
fileName, fileUrl, fileMegabytesThe file that was transcribed
durationSeconds, durationLength of the audio
language, languageConfidenceThe language used, and how sure the detection was (empty when you gave the language)
textThe full transcript
segmentsEvery spoken segment with start and end in seconds and its text
wordCountWords in the transcript (empty for languages written without spaces)
txtUrl, srtUrl, vttUrl, jsonUrlLinks to the transcript as plain text, SRT subtitles, WebVTT subtitles and JSON
quality, modelThe quality chosen and the Whisper model used
minutesChargedAudio minutes charged for this file: its length rounded up to a full minute
recordId, transcribedAtA stable ID for the file and quality, and when it was transcribed

The run report (REPORT, a page you can open, and SUMMARY, the same as JSON) lists every link with what happened to it and why anything was skipped.

What it can and cannot do

  • Whisper writes what it hears and can mishear names, numbers and overlapping speech. Best and Accurate make fewer such mistakes than Fast; none is perfect. Check anything you will quote.
  • It does not tell speakers apart. The transcript has no speaker names.
  • On Fast and Accurate one language is used for a whole file, and a part in a second language comes out poorly. Best detects the language for each 90 seconds or so, so it handles recordings that switch language better.
  • Music, silence and noise without speech give no transcript and are not charged.
  • Video is transcribed from its sound track. Text shown on screen is not read.
  • Files up to 4 GB and 10 hours can be set in the limits; a file must also fit in the run's timeout. The run says so before starting a file it cannot finish.
  • On Fast and Accurate, speed depends on the run's memory: 8 GB is about twice as fast as 4 GB at the same cost per minute.

Pricing

EventPrice
Audio minute, Fast or Best$0.006
Audio minute, Accurate$0.015

A file is charged for its length rounded up to a full minute, once its transcript is saved. Apify platform usage is included. Paid Apify plans get a discount (Bronze 10%, Silver 15%, Gold and higher 20%).

Examples:

  • A 10-minute voice memo on Fast: $0.06.
  • A 45-minute meeting recording on Fast: $0.27.
  • A 2-hour lecture on Best: $0.72; on Accurate: $1.80.

Not charged: files that cannot be downloaded or opened, files with no speech, files over your limits, and files the run has no time left for.

The run stops at the Maximum cost per run you set, before starting a file it could not pay for, and says so.

Common questions

Is my audio sent to OpenAI or another service?

Never to OpenAI. On Fast and Accurate, the Whisper model is part of the Actor and runs on Apify's servers during your run, and the audio is not sent anywhere else. On Best, the audio goes in short pieces to Cloudflare Workers AI, which runs Whisper; Cloudflare says it doesn't keep what it processes or train on it. The downloaded file and its decoded audio are deleted when the file is done.

Which languages work?

About 100, the languages Whisper knows. Common ones (English, Spanish, German, French, Portuguese, Japanese, Chinese and others) work well on Fast. For less common languages choose Best, or Accurate if the audio must stay on Apify, and give the language.

Can I get subtitles?

Yes. Every file gets an SRT and a VTT file with timestamps, linked from its row.

Can I export the results to Excel or CSV?

Yes. Open the Output tab and choose Export, or fetch the dataset through the API in CSV, Excel, JSON or XML.

Can I transcribe a file from my computer?

Put it somewhere with a link: Dropbox or Google Drive shared with anyone who has the link, or your own cloud storage. Paste that link.

Why was a file skipped?

The run report gives the reason for each link: not a media file, the host refused the download, longer or larger than your limits, no speech found, or not enough time left in the run.

Can I transcribe YouTube videos or podcasts?

Not with this Actor. It needs links to the files themselves. For podcasts use Audio Transcription and Speech to Text for Podcasts.

Support

Questions or a result that looks wrong: open an issue on the Issues tab or email contact.cleanscrape@gmail.com. Include the run ID.

If the Actor was useful, an honest review on its Store page helps other people find it.

Whisper is an open speech recognition model released by OpenAI. This Actor is an independent tool and is not affiliated with, endorsed by or sponsored by OpenAI or Cloudflare.

Maintained by CleanScrape.

More tools from CleanScrape