Audio & Video Transcription (Whisper): Podcast, MP3 to SRT

Audio & Video Transcription with Whisper: podcasts, MP3, MP4 to text and SRT

Transcribe audio and video files, podcast RSS feeds and Apple Podcasts links to text with OpenAI's open-source Whisper model. Each file gives you plain text, timestamped segments, SRT and VTT subtitles and, if you want them, word-level timestamps, in 99 languages with automatic language detection.

• Any direct media link. MP3, M4A, WAV, FLAC, OGG, OPUS, AAC, MP4, MOV, WEBM, MKV, AVI and more. Video files are fine: only the audio track is read.
• Podcasts built in. Paste an RSS feed and choose how many episodes, with title and date filters. Apple Podcasts show and episode links are resolved to the show's public feed. Episode title, date and podcast name are added to every row.
• Subtitles that read well. SRT and VTT cues are cut from word timings at sentence and clause boundaries, at most 42 characters per line and 2 lines, which is the broadcast standard (you can change both).
• Private by design. Transcription runs inside the Actor with faster-whisper. Your audio is not sent to OpenAI or any other third-party API.
• Pay per audio minute. From $0.004 per minute. Failed downloads, broken files and files with no speech are free.

What can you use it for?

Use case	How
Podcast show notes, blog posts and SEO	Give the feed URL, set maxEpisodesPerFeed, then use text
Subtitles for videos and reels	Give the MP4/MOV link or upload it, then download srt_url or vtt_url
Searchable archives of interviews, lectures and meetings	Use segments (sentence + start/end seconds) to jump to the moment
AI agents, RAG and LLM pipelines	Feed text or segments to Claude, GPT or a vector store
Translate any language to English	task: translate
n8n / Make / Zapier automations	Call the Actor with a file URL and read the dataset row

Input

{
  "mediaUrls": [
    "https://feeds.example.com/my-podcast.rss",
    "https://cdn.example.com/interview.mp3",
    "https://example.com/videos/webinar.mp4"
  ],
  "maxEpisodesPerFeed": 3,
  "model": "small",
  "language": "auto",
  "wordTimestamps": false
}

Field	Default	What it does
mediaUrls		Direct audio/video file links, podcast RSS/Atom feeds, Apple Podcasts links, or web pages that embed a media file (<audio>, <video>, og:audio). Google Drive and Dropbox share links work
uploadedFiles		Upload files from your computer in the Console
maxEpisodesPerFeed	1	Episodes per feed, newest first. 0 = all that match
episodeTitleFilter		Only episodes whose title contains this text. Use | for alternatives
episodesPublishedAfter, episodesPublishedBefore		YYYY-MM-DD, or relative like 30 days
model	base	tiny, base, small, medium or large-v3-turbo (see below)
language	auto	Spoken language, or auto-detect
task	transcribe	translate gives an English translation
wordTimestamps	off	Every word with start, end and confidence. No extra cost
initialPrompt		Names and jargon to steer spelling, e.g. "Kubernetes, Grafana, Dr. Nkemelu"
vadFilter	on	Skip silence and music (faster, avoids made-up text in silent parts)
subtitleMaxLineLength, subtitleMaxLines	42 / 2	Subtitle layout
maxDurationMinutes	whole file	Transcribe only the first N minutes of each file
maxFileSizeMb, downloadTimeoutSecs	2000 / 900	Limits
httpHeaders		Headers for private file servers or private feeds

Which model?

Measured on Apify at the default 4 GB of memory (1 CPU core), on a 5.4-minute LibriVox recording:

Model	Speed	Accuracy	Price per minute
tiny	about 20x realtime	Rough drafts, keyword search	$0.004
base (default)	about 8x realtime	Good for clear English speech	$0.004
small	about 3x realtime	Clearly better on names, accents and other languages	$0.009
medium	about 1x realtime (2x at 8 GB)	High accuracy	$0.025
large-v3-turbo	about 1x realtime (2x at 8 GB)	High accuracy, best for many languages	$0.025

A 1-hour podcast takes about 8 minutes with base and about 20 minutes with small at 4 GB. More memory makes runs faster: Apify gives one CPU core per 4 GB, and the Actor uses all of them. For medium and large-v3-turbo, use 8-16 GB. The price per minute is the same at any memory.

Output

One dataset row per file. The TXT, SRT, VTT and JSON files are also saved to the run's key-value store and linked from the row.

{
  "media_id": "media-0001",
  "status": "succeeded",
  "source_url": "https://feeds.example.com/my-podcast.rss",
  "source_type": "rss",
  "media_url": "https://cdn.example.com/episodes/ep42.mp3",
  "title": "Ep 42: Pricing your first product",
  "podcast_title": "The Indie Hour",
  "episode_published_at": "2026-11-03T06:00:00+00:00",
  "duration_secs": 2843.2,
  "transcribed_secs": 2843.2,
  "billed_minutes": 48,
  "model": "small",
  "language": "en",
  "language_probability": 0.998,
  "text": "Welcome back to the Indie Hour. Today we're talking about pricing...\n\nSo the first thing I did was...",
  "word_count": 7412,
  "segments": [
    { "id": 0, "start": 0.42, "end": 3.1, "text": "Welcome back to the Indie Hour." },
    { "id": 1, "start": 3.48, "end": 7.92, "text": "Today we're talking about pricing your first product." }
  ],
  "srt_url": "https://api.apify.com/v2/key-value-stores/.../records/media-0001-ep42.srt",
  "vtt_url": "https://api.apify.com/v2/key-value-stores/.../records/media-0001-ep42.vtt",
  "text_url": "https://api.apify.com/v2/key-value-stores/.../records/media-0001-ep42.txt",
  "json_url": "https://api.apify.com/v2/key-value-stores/.../records/media-0001-ep42.json",
  "processing_secs": 951.3,
  "speed_x": 3.0,
  "errors": [],
  "warnings": []
}

With wordTimestamps on, rows also have words: [{"start": 0.42, "end": 0.71, "word": " Welcome", "probability": 0.98}, ...].

SRT example:

1
00:00:00,420 --> 00:00:03,100
Welcome back to the Indie Hour.

2
00:00:03,480 --> 00:00:07,920
Today we're talking about pricing
your first product.

Failed files get status: "failed" and an error code, such as download_failed, unsupported_platform, not_media, no_media_found, no_audio_stream, decode_failed, no_episodes or file_too_large. Files skipped because the run hit your cost limit get status: "skipped". Neither is charged.

The dataset has four views: Overview, Transcripts, Segments (one row per sentence) and Status & errors.

Use from the API, n8n or Make

curl -X POST "https://api.apify.com/v2/acts/tidewell~media-transcriber/run-sync-get-dataset-items?token=<APIFY_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{"mediaUrls": ["https://example.com/interview.mp3"], "model": "small"}'

Synchronous calls wait up to 5 minutes, which is enough for about 30 minutes of audio with base. For longer files, start the run asynchronously and read the dataset when it finishes. In n8n or Make, use the Apify node with "Run Actor" and "Get dataset items".

Python:

from apify_client import ApifyClient

client = ApifyClient("<APIFY_TOKEN>")
run = client.actor("tidewell/media-transcriber").call(run_input={
    "mediaUrls": ["https://feeds.example.com/my-podcast.rss"], "maxEpisodesPerFeed": 5, "model": "small",
})
for row in client.dataset(run["defaultDatasetId"]).iterate_items():
    if row["status"] == "succeeded":
        print(row["title"], row["srt_url"])

Pricing

Pay per event, per minute of audio transcribed. A file is billed seconds transcribed / 60, rounded up to the next minute.

Event	Price	When
Audio minute (tiny or base)	$0.004	Per minute transcribed with tiny or base
Audio minute (small)	$0.009	Per minute transcribed with small
Audio minute (medium or large-v3-turbo)	$0.025	Per minute transcribed with medium or large-v3-turbo
Actor start	$0.0005 per GB of memory	Once per run ($0.002 at the default 4 GB)

Example	Cost
1-hour podcast episode, base	$0.24
1-hour podcast episode, small	$0.54
10 x 45-minute episodes, small	$4.05
100 one-minute clips, base (one run)	$0.40

Failed downloads, broken or unsupported files and files with no speech cost nothing. The Actor respects your maximum cost per run: before each file it works out how many minutes your limit still pays for and transcribes only that much, so it never does work it cannot bill.

FAQ

Can I transcribe YouTube, TikTok, Instagram, Spotify or Vimeo links? No. Downloading from those platforms breaks their terms of service, and Spotify audio is DRM-protected, so the Actor refuses those links with unsupported_platform (not charged). If you own the video, download it and upload the file, or use a direct file link from your own storage. For YouTube captions, a YouTube transcript Actor that reads the official captions is the better tool.

Which podcast links work? Any public RSS or Atom feed with audio/video enclosures (Buzzsprout, Libsyn, Anchor/Spotify for Podcasters RSS, Megaphone, Transistor, Podbean, Simplecast and more), and Apple Podcasts show or episode links, which are resolved through Apple's public lookup API to the show's own feed. Spotify-exclusive shows have no public feed.

Does it identify speakers (diarization)? Not yet. You get timestamps per sentence and per word, not speaker labels.

How accurate is it? It is Whisper. base is good on clear speech in English and the big European languages. Use small or larger for accents, names, noisy audio, phone calls and other languages. Adding names and jargon to initialPrompt helps with spelling.

What about music, silence or ads with no speech? Voice activity detection skips them, so they take little time and produce no made-up text. A file with no speech at all is not charged.

How long can files be? There is no fixed limit. Audio is decoded in 10-minute windows, so memory stays low even for multi-hour recordings. Very long files simply take longer; use more memory to speed them up. Files over 2 GB need maxFileSizeMb.

Is my audio stored or shared? Files are downloaded to the Actor's temporary disk and deleted after transcription. Transcripts are saved only to your own run's dataset and key-value store, under your retention settings. Nothing is sent to OpenAI or other APIs.

Are the transcripts mine to use? The transcript is yours, but the rights in the audio belong to whoever owns it. Only transcribe media you own or have permission to use.

Keywords

audio to text, speech to text, transcribe mp3, mp3 to text, video to text, mp4 to text, podcast transcription, podcast transcript, rss to text, whisper, whisper api, faster-whisper, openai whisper, srt subtitles, vtt subtitles, subtitle generator, word timestamps, transcription api, transcribe audio, audio transcription, wav to text, m4a to text, translate audio to english