DEV Community

Cover image for YouTube transcript APIs: existing captions or speech-to-text
FlashData
FlashData

Posted on Originally published at flashdata.dev

YouTube transcript APIs: existing captions or speech-to-text

Choose an existing-caption API when you need the text and timestamps already published on YouTube. Choose a service with documented speech-to-text fallback when videos without captions must still yield text, and budget for minutes rather than assuming one cheap request. Official YouTube captions require video-management permission; a self-hosted library shifts operations to you. The hosted APIs below differ in both missing-track behavior and billing.

Provider Text source and timestamps Entry price and unit Free or trial When to choose it
YouTube Data API captions (official) Authorized caption-track download; the caller must be allowed to edit the video. Timed subtitle formats, not arbitrary public-video access or STT. 200 quota units per captions.download call; not a USD-per-1,000 tariff. Source · 2026-11-15 UTC Uses the project’s API quota; not a transcript trial package. Owners or authorized managers downloading their own caption tracks.
youtube-transcript-api (jdepoix) Self-hosted Python library reading existing manual or YouTube-generated tracks; timed snippets. No hosted STT fallback. Open-source software: $0 license fee; your compute and any proxy costs are separate. Not directly comparable to hosted requests. Source · 2026-11-15 UTC No SaaS trial; run it yourself. Teams comfortable operating extraction and handling blocked cloud IPs.
Supadata Existing transcripts with timestamps; AI speech recognition fallback when captions are absent. Basic: $5/month equivalent, billed $60/year, 300 credits/month. Existing transcript: 1 credit; generated: 2 credits/video minute. Source · 2026-11-15 UTC Free: 100 credits/month; generated minutes consume that budget faster than existing-track requests. Applications that need text even when no caption track exists and can budget for STT.
TranscriptAPI Transcript JSON with or without timestamps. Public pages checked do not establish whether missing captions trigger STT. Monthly: $5 for 1,000 credits; 1 transcript = 1 successful request = 1 credit. Source · 2026-11-15 UTC 100 free trial credits, no card required. A simple successful-request allowance; confirm missing-caption behavior if it is essential.
Captapi Existing captions: its FAQ says missing tracks return 404 no_captions, charged 0 credits. Transcript timestamp fields were not established by the public pages checked. Starter: $9/month for 2,000 credits. Transcript-specific credits/request not verified; do not read this as 2,000 transcripts. Source · 2026-11-15 UTC 100 lifetime credits after email verification; not a monthly refill. Teams using its wider API catalog and accepting an explicit no-captions failure.
FlashData YouTube Data Transcript defaults to YouTube automatic captions; Captions defaults to uploader tracks. Both return timed text. No generated transcription, translation or source fallback. PAYG: $0.50/1,000 standard requests; no monthly fee. Source · 2026-11-15 UTC Shared new-account trial credits; failed requests are not billed. Applications needing existing tracks with explicit source selection and timestamp segments.

FlashData’s Google Search and YouTube Download are separate products sharing the account key. Download bytes are not the billing unit for the YouTube Data row above.

An automatic YouTube caption is still an existing track

“Automatic” can describe text YouTube has already generated, or a provider generating new text from audio now. These are different operations. The jdepoix library reads available manual or generated tracks and supports YouTube-provided translations. Google’s authorized captions.download also has a translation parameter. Neither fact means FlashData translates or generates a missing track.

For FlashData, Transcript selects automatic captions by default and Captions selects uploader captions by default. The auto parameter explicitly selects the source; the API does not silently switch sources. Both return timestamp segments and full text. If the requested track or language does not exist, there is no STT fallback.

When missing captions must still produce text

Supadata explicitly documents AI speech recognition through the same endpoint when captions are absent. Its generated-transcript unit is video minutes, while existing-transcript retrieval uses credits per request. Its pricing page gives a 100-credit monthly free budget; a product-page phrase of “100 requests/month” should not be read as 100 arbitrarily long generated transcripts.

Captapi documents a no_captions error rather than an automatic transcription fallback. TranscriptAPI’s public pricing and examples establish timestamps and one-credit successful requests, but did not establish its no-caption policy in this review. If this behavior determines your choice, ask for the endpoint’s documented behavior instead of inferring it from “transcript API.”

Keep quota, request fees and usage rights separate

The official YouTube captions API requires OAuth authorization and permission to edit the video. Its 200 quota units are not a dollar rate and cannot be converted into a hosted API’s price per 1,000 requests without inventing an exchange rate. For self-hosted youtube-transcript-api, operating costs and cloud-IP blocks belong to the caller.

FlashData Extended Search has a different rate from standard YouTube Data requests and is unrelated to retrieving a transcript. A license check reports whether a video is labeled Creative Commons; it does not grant training rights. Trial credits are shared across the account. A job marked failed is not billed; a client timeout alone does not prove that the job failed.

Implementation after selection

For FlashData, read the Supadata comparison, JavaScript transcript tutorial, caption-language guide, Transcript reference and Captions reference.

Where the numbers come from

All prices are in USD. Official pages were checked on 2026-11-15 UTC. This is a vendor-document comparison published by FlashData, not an independent performance benchmark.

Originally published at FlashData.

Top comments (0)