Podcast & Audio Transcriber: MP3, Video to Text

Turn podcasts, MP3s and video files into text with OpenAI's open Whisper speech model, run on Apify: no API key, no OpenAI account, no software to install. Give it a link to an audio or video file, a podcast RSS feed, an Apple Podcasts page or just the show's name, or upload a file, and get the transcript with timed segments and SRT or WebVTT subtitles, in 99 languages. It is built for podcasters, researchers and journalists, marketers who repurpose episodes, and AI agents that need what was said in an episode as text.

• Podcasts by name or feed. It finds the show, picks the newest episodes or those matching a date or title, and on a schedule transcribes each new episode once.
• The publisher's own transcript when there is one. Many feeds link a transcript made by the show. The Actor uses it for a flat $0.005 an episode instead of transcribing the audio.
• Pay per audio minute, $10 per 1,000 minutes ($7.50 on Gold), with nothing to pay for silent files, failed links or the part of a file you cut off.

The easiest way to try it: keep the example input and press Start. You get the first two minutes of the newest Planet Money episode as text and SRT subtitles in under a minute.

What one transcript looks like

{
    "showTitle": "Planet Money",
    "episodeTitle": "Wanna slow down AI? These biologists can help.",
    "publishedAt": "2026-10-14T21:30:18.000Z",
    "durationSeconds": 1957,
    "transcribedSeconds": 120,
    "transcriptSource": "whisper",
    "model": "base",
    "language": "en",
    "text": "… There's a very strange puzzle at the center of the AI debate right now. …",
    "segments": [
        { "start": 30.33, "end": 33.05, "text": "- There's a very strange puzzle", "speaker": null },
        { "start": 33.05, "end": 35.37, "text": "at the center of the AI debate right now.", "speaker": null }
    ],
    "wordCount": 353,
    "billedEvent": "audio-minute",
    "billedCount": 2
}

These are 13 of the 29 fields of a result; full results are under Output below.

What you can do with it

• Make a podcast searchable. Transcribe a show's back catalogue once, then schedule a weekly run with Only episodes not transcribed before to add each new episode.
• Write show notes, articles and social posts from an episode, or feed the text to an LLM for summaries and quotes.
• Add subtitles to a video. The SRT and WebVTT files load into YouTube Studio, video players and editors.
• Research and monitoring. Follow what a set of shows says about a company, product or topic.
• Give an AI agent or a spreadsheet what was said, in one fixed format with timestamps.

How to use it

1. Add your audio under Links to audio, video or podcasts (one per line), Podcast names or Upload files. You can mix them in one run.
   • A direct file link: MP3, M4A, WAV, OGG, OPUS, FLAC, AAC, MP4, M4V, WEBM, MOV and other formats ffmpeg reads.
   • A podcast RSS feed, an Apple Podcasts show page (podcasts.apple.com/…/id1200361736) or episode page (…?i=1000…), or a web page that embeds an audio file or names its feed.
   • A show name such as Planet Money, looked up in the Apple Podcasts directory.
2. For podcasts, choose how many Episodes per podcast (newest first), and optionally Published after a date or period such as 30 days, or Episode title contains.
3. Choose the Speech model (base by default), the Spoken language (auto-detect by default) and whether to keep the language or translate to English. Add Subtitle files if you want SRT or WebVTT.
4. Set Maximum minutes per file or Maximum minutes per run if you only need the start of each file or want to cap the cost. Leave them empty to transcribe everything.
5. Press Start, then export the results as JSON, CSV or Excel, or read them through the API.

To transcribe every new episode of a show automatically, save the input as a task with Only episodes not transcribed before switched on, and give the task a daily or weekly schedule. Each run looks at the newest episodes, skips those your earlier runs already transcribed and costs only the run start when nothing is new.

How long it takes

Measured on Apify on 2026-10-16 with the base model:

Memory	Whisper workers	Speed	A 60-minute episode
4 GB (default)	1	about 1 minute of audio every 13 to 15 seconds	about 14 minutes
8 GB	2	32.4 minutes of audio in 4 minutes 7 seconds	about 7 minutes
16 GB	4	four parts at once (not yet timed)	about 4 minutes

A run spends 15 to 25 seconds starting before the first file. small takes about three times as long as base (10 minutes of audio in 6 minutes 21 seconds at 4 GB), tiny less. More memory costs the same per audio minute, only the run start grows by $0.00005 per GB, so give a long episode or a large batch 8 or 16 GB. The default timeout of 3 hours holds about 800 minutes of audio at 4 GB with base; raise the memory or the timeout for more.

Pricing

You pay per minute of audio transcribed, or a flat price per episode when the publisher's transcript is used, plus Apify's run start of $0.00005 per GB of memory ($0.0002 at the default 4 GB). Platform usage is included, so there is nothing else to pay.

Apify plan tier	Audio minute, tiny or base model	Audio minute, small model	Publisher transcript
Free, Bronze	$0.01	$0.025	$0.005
Silver	$0.0088	$0.022	$0.0044
Gold and above	$0.0075	$0.01875	$0.00375

Minutes are counted per file and rounded up: a 30-second clip counts as one minute, a 45 min 10 s episode as 46. Worked examples at the Free and Bronze price:

• The example input, the first 2 minutes of one episode: 2 x $0.01 = $0.02.
• A 60-minute episode with the base model: $0.60; with the small model, $1.50.
• 1,000 minutes of audio: $10.00, or $8.80 on Silver and $7.50 on Gold.
• A weekly 45-minute show on a schedule: $0.45 a week, about $1.95 a month.
• An episode whose feed links a transcript: $0.005, whatever its length.
• A silent file, a file without words, or a link that fails: nothing.

Maximum minutes per file and Maximum minutes per run cap what a run transcribes, and only the minutes transcribed are charged. You can also set a maximum charge per run: the Actor transcribes only as many minutes as that amount pays for, cuts the last file where the money ends, and finishes as succeeded with a message that says the limit stopped it. A run that Apify restarts on another server carries on where it stopped and charges no file twice.

Input

{
    "mediaUrls": [
        "https://feeds.npr.org/510289/podcast.xml",
        "https://podcasts.apple.com/us/podcast/the-daily/id1200361736",
        "https://example.com/interviews/2026-10-01.mp3"
    ],
    "podcastNames": ["Freakonomics Radio"],
    "maxEpisodesPerFeed": 5,
    "publishedAfter": "30 days",
    "onlyNewEpisodes": false,
    "usePublisherTranscripts": true,
    "model": "base",
    "language": "auto",
    "task": "transcribe",
    "subtitleFormats": ["srt", "vtt"],
    "maxMinutesPerFile": 90
}

Field	What it does
mediaUrls	Files, feeds, Apple Podcasts pages or web pages, one per line.
podcastNames	Show names to look up in the Apple Podcasts directory.
audioFiles	Files you upload, up to 1.5 GB each.
maxEpisodesPerFeed	Episodes per feed, show or name, newest first. Default 3.
publishedAfter	Only episodes published on or after a date (2026-01-31) or period (30 days).
titleContains	Only episodes whose title contains this text.
onlyNewEpisodes	Skip episodes an earlier run in your account transcribed. For schedules.
usePublisherTranscripts	Use the feed's own transcript when it links one. Default on.
model	tiny, base (default) or small.
language	auto or one of 99 language codes.
task	transcribe keeps the language; translate gives English text.
subtitleFormats	Add srt, vtt or both to each result.
maxMinutesPerFile	Transcribe at most this many minutes from the start of each file.
maxTotalMinutes	Stop once the run has transcribed this many minutes.

Output

One result per file or episode, in the order of your input. A transcribed episode, a publisher transcript and a link that failed:

[
    {
        "inputUrl": "https://feeds.npr.org/510289/podcast.xml",
        "sourceType": "feed",
        "status": "transcribed",
        "showTitle": "Planet Money",
        "feedUrl": "https://feeds.npr.org/510289/podcast.xml",
        "episodeTitle": "Wanna slow down AI? These biologists can help.",
        "episodeGuid": "b297c28a-87e4-4b96-92e0-e64b7b581948",
        "publishedAt": "2026-10-14T21:30:18.000Z",
        "mediaUrl": "https://tracking.swap.fm/track/…/default.mp3",
        "durationSeconds": 1957,
        "transcribedSeconds": 120,
        "truncated": true,
        "truncatedReason": "maxMinutesPerFile",
        "transcriptSource": "whisper",
        "publisherTranscriptFormat": null,
        "model": "base",
        "language": "en",
        "languageProbability": 0.999,
        "task": "transcribe",
        "text": "… There's a very strange puzzle at the center of the AI debate right now. …",
        "segments": [
            { "start": 30.33, "end": 33.05, "text": "- There's a very strange puzzle", "speaker": null },
            { "start": 33.05, "end": 35.37, "text": "at the center of the AI debate right now.", "speaker": null }
        ],
        "srt": "…\n12\n00:00:30,330 --> 00:00:33,050\n- There's a very strange puzzle\n…",
        "vtt": null,
        "wordCount": 353,
        "billedEvent": "audio-minute",
        "billedCount": 2,
        "error": null,
        "errorKind": null,
        "transcribedAt": "2026-10-16T12:00:18.578Z"
    },
    {
        "inputUrl": "https://feeds.megaphone.fm/darknetdiaries",
        "sourceType": "feed",
        "status": "transcribed",
        "showTitle": "Darknet Diaries",
        "feedUrl": "https://feeds.megaphone.fm/darknetdiaries",
        "episodeTitle": "180: Conti",
        "episodeGuid": "prx_7057_141e0c43-f935-41de-be94-708448052c4d",
        "publishedAt": "2026-10-06T07:00:00.000Z",
        "mediaUrl": "https://www.podtrac.com/pts/redirect.mp3/dovetail.prxu.org/7057/141e0c43-f935-41de-be94-708448052c4d/darknet-diaries-ep180-conti.mp3",
        "durationSeconds": 3557,
        "transcribedSeconds": null,
        "truncated": false,
        "truncatedReason": null,
        "transcriptSource": "publisher",
        "publisherTranscriptFormat": "vtt",
        "model": null,
        "language": null,
        "languageProbability": null,
        "task": null,
        "text": "[Music] There’s a kind of silence that follows a ransomware attack. …",
        "segments": [
            { "start": 0, "end": 1.18, "text": "[Music] There’s a kind of silence", "speaker": "JACK" },
            { "start": 1.18, "end": 3.22, "text": "that follows a ransomware attack.", "speaker": null }
        ],
        "srt": "1\n00:00:00,000 --> 00:00:01,180\nJACK: [Music] There’s a kind of silence\n…",
        "vtt": null,
        "wordCount": 10188,
        "billedEvent": "publisher-transcript",
        "billedCount": 1,
        "error": null,
        "errorKind": null,
        "transcribedAt": "2026-10-16T12:01:41.314Z"
    },
    {
        "inputUrl": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
        "sourceType": "file",
        "status": "failed",
        "text": "",
        "segments": [],
        "wordCount": 0,
        "billedEvent": null,
        "billedCount": 0,
        "error": "YouTube links are not supported: this Actor transcribes audio and video files and podcast feeds. Use a direct link to the file, the show's RSS feed, or upload the file.",
        "errorKind": "unsupported-platform"
    }
]

(The failed result is shortened here; every result carries all 29 fields.)

• status: transcribed, no-speech (nothing to transcribe, not charged) or failed.
• transcriptSource: whisper when the audio was transcribed, publisher when the feed's transcript was used.
• segments: phrases of a few seconds with start and end in seconds from the start of the file. speaker is filled only when a publisher transcript names the speakers.
• truncated and truncatedReason: whether a limit cut the file, and which one.
• billedEvent and billedCount: what this result cost, so you can check every charge.
• errorKind: unsupported-platform, not-found, refused, outage, too-large, not-media, no-audio, undecodable, no-public-feed, episode-not-found, invalid-input, engine or error.

Limits

• No YouTube, TikTok, Instagram, Spotify or other platform pages. Those links fail with unsupported-platform and cost nothing. Use the show's RSS feed, a direct file link or an upload.
• Only audio that opens without a login. Private, members-only and paywalled feeds are not supported.
• No speaker labels from Whisper. Speakers appear only when the publisher's transcript names them.
• No word-level timestamps, custom vocabulary, summaries or chapters. Names and jargon can come out misspelt; segments are phrases, not single words.
• Translation goes to English only, and publisher transcripts are not used when translating.
• Accuracy follows the audio. Clear speech comes out well with base; noisy recordings, heavy accents, crosstalk and music do better with small and can still contain mistakes. Like every Whisper system it can, rarely, repeat a phrase or write words over music.
• The whole file is downloaded, even when only its first minutes are transcribed. Files up to 1.5 GB.
• Publisher transcripts are used as published. Their quality, and any missing language, is the publisher's.

FAQ

Is it legal to transcribe podcasts?

The Actor reads only public feeds, pages and files that open without a login, and it does not get around any blocking. Copyright in an episode stays with its publisher: transcripts for research, search, accessibility, notes or quoting are common, while republishing whole transcripts usually needs the publisher's permission. Check the terms of the show and the laws that apply to you; this is not legal advice.

Why did a link fail?

The error field says why in plain words and errorKind gives the cause. The usual ones: the link is a platform page (unsupported-platform), the server answered 404 (not-found) or refused the download (refused), the file has no audio track (no-audio), or a page links no audio file and no feed (no-public-feed). A failed link is never charged, and the other links in the run carry on.

How do I transcribe only new episodes?

Switch on Only episodes not transcribed before, save the input as a task and schedule it. The Actor remembers which episodes your account transcribed. It looks at the newest episodes per feed (Episodes per podcast) and never works back through the archive, so set that number to at least the episodes a show releases between two runs.

Which languages does it understand?

The 99 languages Whisper knows, from English, Spanish and German to Japanese, Arabic and Hindi. Auto-detect listens to two stretches of each file; set the language when you know it, especially for short files. translate turns any of them into English text.

Which model should I choose?

base suits most clear speech at $0.01 a minute. small is more accurate on noisy audio, accents and less common languages, at $0.025 a minute and about three times the time. tiny is the fastest and roughest, at the base price.

Can I use it through the API or from an AI agent?

Yes. Start it through the Apify API, the Python or JavaScript client, Make, Zapier or n8n, or call it from an agent through the Apify MCP server. Every result has the same fields.

More Actors from this developer

• ATS Job Postings: Workday, Greenhouse, Lever & Ashby: every open job of the companies you choose, with only the postings that are new since the last run if you want.
• Company Jobs Search: open jobs by title on 1,300 company career sites in one run.
• Website Contact Scraper: emails, phones and social profiles from company websites.
• Email Validator & List Cleaner: syntax and domain checks for a mailing list.
• Handelsregister Scraper and European Company Registry Search: company records from official registers.