Audio & Podcast Transcriber (Speech to Text)

Podcast and audio transcription with timestamps. Add direct links to MP3, M4A, WAV, OGG, Opus or MP4 files, or a podcast RSS feed, and get the text, plus SRT or VTT subtitles if you want them.

No login and no API key. The Actor runs the open-source Whisper speech-to-text model through faster-whisper on Apify, so your audio is not sent to a third-party AI service. A 1-minute clip is ready in about 30 seconds and costs about $0.03.

Table of contents

• What it does
• Quick start
• Input
• Output
• Pricing
• Use cases
• Scheduling
• API, export and integrations
• Limits and partial results
• Troubleshooting
• FAQ
• Related Actors
• Support

What it does

The Actor downloads each media file, reads the audio track and transcribes it with Whisper. You get one dataset row per file or podcast episode. Each row has the full text, the language, the duration and, if you choose, timed segments or a subtitle file.

• Input: direct media links (audioUrls), or a podcast RSS feed or Apple Podcasts show link (podcastRssUrl).
• Output: the transcript text, timed segments {start, end, text}, SRT or WebVTT subtitles, the detected language and the episode details from the feed.
• Languages: Whisper detects the spoken language. You can also choose it, or translate the speech to English.
• Privacy: the audio is processed inside the run and is not saved to storage.

Good fit

• Transcribe the newest episodes of a podcast from its RSS feed.
• Make SRT or VTT subtitles for a video file you host.
• Turn interviews, lectures and voice notes into searchable text for notes, search or an AI pipeline.

Not a fit / limits

• No downloads from YouTube, TikTok, Spotify, Instagram or other page links. The link must point straight to a media file (the URL usually ends with .mp3, .m4a, .wav, .ogg, .mp4 and so on).
• No speaker labels (who said what). There is one text stream per file.
• Live streams are not supported.
• Whisper can misspell names and rare words. Check important quotes against the audio.

Quick start

1. Open the Input tab.
2. In Audio or video URLs (audioUrls), keep the example LibriVox file or paste your own direct media link.
3. Keep the defaults: model base and output segments.
4. Click Start.
5. Open the Transcripts view in the Output tab and read the text column.

Expected result: one row with status complete, the language en and about 160 words of text. It takes about 25–30 seconds.

Input

Input fields

Field	Plain name	Type	Required	Default / prefill	Allowed values	Effect	Recommendation
audioUrls	Audio or video URLs	array of strings	one of audioUrls or podcastRssUrl	prefill: 1 LibriVox MP3 (78 s)	up to 100 public http(s) links to media files	each link becomes one row	use direct file links; duplicates are removed
podcastRssUrl	Podcast RSS feed	string	one of audioUrls or podcastRssUrl	empty	RSS feed URL or podcasts.apple.com/.../id123 link	adds the newest episodes of the feed	use for podcasts; Apple links are turned into the RSS feed
maxEpisodes	Max episodes	integer	no	3	1–50	how many newest feed episodes to transcribe	start with 1–3
model	Model	string	no	base	tiny, base, small	speed, quality and price	base for most audio; tiny for fast drafts; small for hard audio and non-English (use 4096 MB memory or more)
language	Language	string	no	auto	auto or a Whisper language code (en, es, de, …)	skips detection when set	set it when you know the language
outputFormat	Output format	string	no	segments	segments, text, srt, vtt	adds timed segments or a subtitle file to each row	srt/vtt for subtitles, text for the smallest rows
translateToEnglish	Translate to English	boolean	no	false	true/false	returns English text instead of the original language	use for non-English audio
maxDurationMinutes	Max minutes per file	integer	no	120	1–180	only the first N minutes of each file are transcribed and charged	lower it to test a long file cheaply

Empty input (no URLs and no feed) stops the run with a clear error. Every row always has text; segments and subtitles depend on outputFormat.

Interacting settings

When	Input A	Input B	Effective behavior
Both are set	audioUrls	podcastRssUrl	both are processed in one run: feed episodes come after the links
Long file	maxDurationMinutes	file length	the file is cut, the row gets status: partial and error code TRUNCATED
Spending cap	Max cost per run in run options	number of minutes	when the cap is reached, the current file stops (partial) and the next files get CHARGE_LIMIT
Language set	language	translateToEnglish	the chosen language is read; with translation the text is in English

Input examples

Newest podcast episodes

{
  "podcastRssUrl": "https://podcasts.apple.com/us/podcast/houston-we-have-a-podcast/id1262594123",
  "maxEpisodes": 2,
  "model": "base",
  "outputFormat": "segments"
}

Expected behavior: the Actor finds the RSS feed of NASA's "Houston We Have a Podcast" from the Apple Podcasts link, takes the 2 newest episodes with audio and transcribes them. Expected output: 2 rows with podcastTitle, title, publishedAt, text and segments.

Subtitles for a video file

{
  "audioUrls": ["https://example.com/media/talk.mp4"],
  "outputFormat": "srt",
  "language": "en"
}

Expected output: 1 row with the SRT file in subtitles.

Output

Output example

Prefill input, outputFormat: segments (segments trimmed):

{
  "sourceUrl": "https://www.archive.org/download/aesop_fables_volume_three_librivox/fables_03_01_aesop_64kb.mp3",
  "sourceType": "url",
  "podcastTitle": null,
  "title": null,
  "publishedAt": null,
  "episodeGuid": null,
  "episodeUrl": null,
  "status": "complete",
  "durationSec": 77.95,
  "transcribedSec": 77.95,
  "truncated": false,
  "language": "en",
  "languageProbability": 0.991,
  "task": "transcribe",
  "model": "base",
  "text": "This is a LibraVox recording. All LibraVox recordings are in the public domain. …",
  "wordCount": 161,
  "segments": [
    {"start": 0.75, "end": 3.65, "text": "This is a LibraVox recording."},
    {"start": 3.65, "end": 6.97, "text": "All LibraVox recordings are in the public domain."}
  ],
  "subtitles": null,
  "chargedMinutes": 2,
  "processingSec": 7.9,
  "error": null,
  "fetchedAt": "2026-10-13T14:43:36Z"
}

The real file says "LibriVox". Whisper wrote "LibraVox": this is a typical name error.

Output fields

Field	Meaning	Type	Can be empty?	Example / enum
sourceUrl	media file URL (your link or the feed enclosure)	string	no	https://…/episode.mp3
sourceType	where the row came from	string	no	url, rss_episode, rss_feed (feed-level error)
podcastTitle	podcast name from the feed	string	null for audioUrls	My Podcast
title	episode title from the feed	string	null for audioUrls	Episode 12
publishedAt	episode date from the feed, ISO 8601 UTC	string	null for audioUrls or when the feed has no date	2026-10-01T08:00:00Z
episodeGuid	episode ID from the feed	string	null for audioUrls	abc-123
episodeUrl	episode web page from the feed	string	null when missing	https://…
status	result of this file	string	no	complete, partial, not_found, error
durationSec	length of the media in seconds	number	null on errors	77.95
transcribedSec	seconds actually transcribed	number	null on errors	77.95
truncated	only the first part was transcribed	boolean	no	false
language	language code used (detected or chosen)	string	null on errors	en
languageProbability	confidence of language detection, 0–1	number	null when you set language or on errors	0.991
task	transcribe or translate	string	no	transcribe
model	Whisper model used	string	no	base
text	full transcript	string	null on errors	This is …
wordCount	words in text	integer	null on errors	161
segments	timed parts {start, end, text} in seconds	array	null unless outputFormat is segments	see example
subtitles	SRT or WebVTT file content	string	null unless outputFormat is srt or vtt	1\n00:00:00,750 --> …
chargedMinutes	minutes charged for this file	integer	0 on errors	2
processingSec	download, decoding and transcription time	number	null when the file was skipped	19.2
error	{code, message} when something went wrong	object	null when none	{"code": "NOT_FOUND", …}
fetchedAt	when the row was made, ISO 8601 UTC	string	no	2026-10-13T14:35:23Z

Output surfaces

Situation	Where	Result	What to do
Normal file	Dataset, view Transcripts	row with status: complete	use text; timed output is in view Timestamps and subtitles
Long file cut by maxDurationMinutes or by the spending cap, or a download that broke off midway	Dataset	status: partial, truncated: true, error code TRUNCATED	raise the limit or the cap and run the file again
File or feed does not exist (HTTP 404), feed without audio episodes	Dataset	status: not_found, code NOT_FOUND or NO_EPISODES	check the link
Broken link, web page instead of a file, YouTube link, slow server	Dataset	status: error with a code	see Troubleshooting
Whole run	Key-value store, record RUN_SUMMARY	status, counts, minutes, error codes	check status and errorCodes

Pricing

Pricing unit

This Actor uses pay-per-event pricing. You pay for:

• Actor start: $0.01 per GB of run memory. The default 2048 MB memory makes it $0.02 per run.
• Transcribed minute (models tiny and base): per started minute of audio in each file.
• Transcribed minute, small model (model small): per started minute of audio in each file.

Event	Free	Bronze	Silver	Gold
Transcribed minute (tiny, base)	$0.0098	$0.0093	$0.0088	$0.0083
Transcribed minute (small)	$0.032	$0.030	$0.029	$0.028

Minutes are counted per file and rounded up: a 78-second file is 2 minutes. Only transcribed audio is charged. Files with status not_found or error cost nothing beyond the start fee. A run where every link is broken costs only the start fee ($0.02 at 2048 MB). Platform usage (compute) is included in these prices.

Cost examples (Free plan, 2048 MB)

Scenario	Audio	Charge
One short clip	1 file × 1 min	$0.030
One podcast episode	1 file × 30 min	$0.314
A long lecture	1 file × 60 min	$0.608
10 files in one run	10 files × 10 min	$1.000
100 files in one run	100 files × 10 min	$9.820
1,000 files in 10 runs	1,000 files × 10 min	$98.20
small model, 4096 MB	1 file × 30 min	$1.000

The start fee is paid once per run, so many files in one run are cheaper than one run per file. 100 files of 10 minutes take about 4 hours with base at 2048 MB: raise the run timeout, or use 4096 MB to make it about 2 times faster at the same price per minute (start fee $0.04). Use Max cost per run in the run options to set a hard spending cap.

Use cases

Buyer job	Input path	Output entity	Key fields	Next step
Podcast show notes and search	podcastRssUrl + maxEpisodes	one row per episode	title, publishedAt, text, segments	send text to a summary or search tool
Subtitles for your video	audioUrls + outputFormat: srt	one row per file	subtitles, language	save subtitles as a .srt file
Meeting or interview notes	audioUrls	one row per file	text, wordCount, durationSec	paste text into notes or an AI agent
English version of foreign audio	audioUrls + translateToEnglish	one row per file	text, language, task	read or index the English text

Scheduling

You can run a podcast feed on an Apify schedule, for example every morning. Each run transcribes the newest maxEpisodes episodes again and charges them again. There is no "only new episodes" mode, so for a daily schedule use maxEpisodes: 1.

API, export and integrations

• API: start the Actor with the Apify API or a client library and read the default dataset.
• Export: download the dataset as JSON, CSV, Excel or HTML from the Output tab or the API.
• Integrations: use Apify integrations or webhooks to send each finished run to Make, Zapier, n8n, Google Drive or your own endpoint.

Use with AI agents (MCP)

• MCP server URL with only this Actor: https://mcp.apify.com?tools=cliqtomedia/audio-video-transcriber.
• The smallest useful input:

{"audioUrls": ["https://www.archive.org/download/aesop_fables_volume_three_librivox/fables_03_01_aesop_64kb.mp3"]}

• Key output fields: status, text, language, segments (with outputFormat: segments), error.code.
• A 1–2 minute file takes about 25–30 seconds and costs about $0.03. A 30-minute file takes about 7 minutes at 2048 MB.
• Only direct media links work, up to 180 minutes per file.

Limits and partial results

Speed

Model	Memory	Time per minute of audio	Note
tiny	2048 MB	about 6 s	fastest, lower quality
base	2048 MB	about 13 s	default
base	4096 MB	about 7 s	same price per minute, start fee $0.04
small	8192 MB	about 12 s	best quality here; give it 4096 MB or more

Limits

• Up to 180 minutes per file (maxDurationMinutes), up to 100 links and 50 feed episodes per run.

• Files up to 1.5 GB.

• The default run timeout is 3600 seconds. A 3-hour file with base at 2048 MB takes about 38 minutes. For several long files, raise the timeout or the memory.

• When the run timeout comes near, the current file is saved as partial and the remaining files get RUN_TIME_LIMIT.

• Partial and failed rows are never charged for audio that was not transcribed. A partial row is charged only for the minutes in its text.

Troubleshooting

The row has error code UNSUPPORTED_SOURCE

Symptom: a YouTube, TikTok, Spotify or similar link returns an error. Reason: these are web pages, not media files. The Actor does not download from them. Action: use a direct link to the media file, for example the MP3 from the podcast RSS feed.

The row has error code NOT_MEDIA

Symptom: the link opens a web page. Reason: the server returned HTML, not audio or video. Action: open the page, find the download link of the file and use it.

The row has error code DOWNLOAD_TIMEOUT or TOO_SLOW

Symptom: the server did not answer for 15 seconds, or sent the file very slowly. Reason: the media server is down or limits downloads. A download slower than about 1 MB/s is checked after a few seconds: if the part received so far is not readable audio, the file stops with TOO_SLOW. After two silent servers, the remaining links in the run are skipped with SKIPPED_SLOW_SOURCES and are not charged. If the connection breaks after part of the file arrived, the Actor transcribes the received part: the row is partial with the reason "the download was interrupted". Action: try again later or use another copy of the file.

The row has status: partial

Symptom: truncated is true and the error code is TRUNCATED, CHARGE_LIMIT or RUN_TIME_LIMIT. Reason: the file is longer than maxDurationMinutes, the run reached its spending cap or timeout, or the download broke off midway (only the received audio is transcribed and charged). Action: raise the limit, the cap or the timeout, then run the file again.

The text has wrong words

Reason: background music, noise, accents or rare names. Action: set language, or try the small model with 4096 MB memory.

FAQ

Do I need an OpenAI or other API key?

No. The Whisper model runs inside the Actor. You only pay the Apify events above.

Which languages work?

Whisper supports about 100 languages. We tested English and Spanish. Quality is best for English and other widely spoken languages; small helps for others.

Are my files stored?

No. The file is downloaded into the run, transcribed and deleted. Only the text rows and the run summary are saved.

Can it tell who is speaking?

No. There are no speaker labels.

Why is a short clip more expensive than one minute?

Each run has a start fee ($0.02 at 2048 MB) and minutes are rounded up per file. Put many files into one run to share the start fee.

Related Actors

• YouTube Subtitles & Transcript Scraper API: get existing subtitles of YouTube videos without speech recognition.
• Apple Podcast Directory & RSS Contacts Scraper: find podcasts and their RSS feeds, then transcribe episodes here.

Support

For bugs, questions or feature requests, open an issue on the Issues tab of this Actor. Add the run ID, the input mode and the error code. Do not share private links or tokens.