Riverside Transcription: Languages, Fixes, Exports and Plan Limits

Riverside transcribes recordings automatically, in one language per recording, with one speaker label per audio track. Riverside says the engine covers 100+ languages at 99% accuracy; treat both as the vendor's figures. Corrections, captions and text-based editing are listed for every plan, but the Free plan's data flags transcriptions off, so do not plan on Free for transcripts. Downloading or copying the transcript as TXT or SRT needs a paid plan, starting with Pro at $35 a month billed monthly or $24 a month billed annually. The Dictionary for names and jargon starts at Grow. Translation into 30 languages runs on AI credits. Our verdict: Pro covers most podcasters and marketers who need clean text out.

Set the studio language before anyone hits record

Riverside transcribes in English by default, and each recording is transcribed in exactly one language: the default language of the studio it was made in. The host can change that setting on any plan, on a computer or in the mobile app (producers, who exist from Grow, can too), but the change applies only to future recordings. A Spanish interview recorded in an English studio therefore comes back as a poor English transcript, and words in a second language may even be translated into the studio language inside the text. If you record in several languages, keep one studio per language and send guests the matching link.

Transcription starts only after every participant's track has uploaded and been processed, so one guest on a slow connection delays the text for everyone. Riverside gives no fixed turnaround; it depends on file size and recording length. An external upload produces a transcript, Magic Clips and editor access on all plans, with a separate audio track per speaker. A mixed video can also be split into per-speaker video tracks, provided it has a fixed layout, a consistent group of participants and no screen sharing. Otherwise you get separate audio only.

What the transcript catches, and what it misses

Speaker labels follow tracks, not voices. Because Riverside records each participant on a separate track, a remote interview gets clean, named speakers with no guessing. Two people sharing one microphone, however, appear under a single name, since the transcript is generated per track. The workaround is to re-upload that file with multiple speakers selected, which creates separate tracks. Renaming a participant in the editor updates the downloadable transcript, the dashboard files and every other edit from the same session. Fix names once, early.

Some audio is never transcribed. Screen-share audio, Media Board sound effects and footage dropped straight into the editor stay out of the text, so a product demo narrated through a shared screen leaves a gap. Filler words are transcribed, and ones the system does not recognize show up as a squiggle marking a disfluency. In non-English recordings, words repeated several times are treated as duplicates and removed, and occasionally a related parent language is picked even when the right language is selected. The mobile app transcribes Hebrew and Arabic but displays them left to right.

Riverside says its transcription is 99% accurate. That is the vendor's number, not ours. Its own help center is more modest: names and similar-sounding terms get misspelled, and quiet speech or crosstalk produces errors. Plan on a proofreading pass for anything published word for word, such as pull quotes, show notes or captions on a sales video. The more guests talk over each other, the longer that pass takes, so build it into the schedule rather than treating the raw transcript as final copy.

Fixing errors with corrections, Correct all and the Dictionary

Corrections are on every plan, in the browser, the mobile apps and the Mac app. Click a word, type the fix, and it flows into the captions and into any transcript or subtitle file you download later. Correct all replaces every occurrence of a word at once, but it is sensitive to case and punctuation, so a name followed by a comma counts as a separate match. There is no bulk delete for words. The help center's workaround is Correct all with a single space as the replacement, which clears every instance in one pass.

Do not regenerate after correcting. Regenerating a transcript replaces it on every edit and clip, and neither manual corrections nor VideoDub fixes carry over. Regeneration does not translate either, so it will not rescue a recording made in the wrong studio language. Deleted text can be shown again and restored, except for parts removed with Split. Words you delete also disappear from the downloaded transcript, while muted words stay in it. If a remark has to vanish for legal reasons, delete it rather than muting it.

The Dictionary prevents errors instead of patching them. You add names, brand terms and jargon, and future transcripts spell them correctly; existing transcripts do not change. Entries are capped at 60 characters, words under three characters are ignored, and participant names are not added automatically, so add guests yourself. It needs Grow, Webinar or Business ^[1]. Grow costs $45 a month billed monthly or $34 a month billed annually. Pro costs $35 a month billed monthly or $24 a month billed annually. The step up is $10 a month on either term, or $120 a year on annual billing, by our arithmetic. If only a handful of names keep breaking, Correct all does the job without the upgrade.

Captions come from the same text

Open captions are listed for all plans, built from the transcript in the studio language, and word timing is most accurate for English. Styling covers presets or custom font, color, size, alignment and animation, and you drag the captions into position on the canvas. Saving caption presets needs Pro or above; presets are unlimited and can become the default for Magic Clips. On any plan you can hide a word or phrase from the on-screen captions without removing it from the transcript, which handles a brand name the engine keeps mangling.

Two limits catch people out. Captions exist only in the language spoken in the recording, original or translated, so you cannot add English subtitles to a Spanish video without translating the video itself. And the SRT file you download leaves out the styling you applied in the editor; it carries text and timing only. Live studio captions are a separate feature. They need Webinar or Business, work in English only, are generated by AI and are not saved to the recording. Riverside cannot switch on captions for viewers of a live stream.

Getting the text out as TXT, SRT or timestamped copy

This is where the plan matters. Downloading the transcript of a full recording, or of an edited one, needs Pro, Grow, Webinar or Business ^[2], and only owners, admins, directors and editors can do it. Files come as plain TXT or as SRT subtitles with a timestamp every few seconds. Copying the transcript with timestamps follows the same plan and role rules, in the browser, the mobile app and the Mac app. Text you cut in the editor is left out of the downloadable file, so export the edit when you want the clean version.

The Free plan is the gap: its plan data flags transcriptions off and download and copy sit on Pro. The cheapest route is Pro at $24 a month billed annually ($288 a year) or $35 a month billed monthly. For a short project, monthly billing wins up to eight months: eight months cost $280 and nine cost $315, against $288 for a year billed annually, by our arithmetic. Plans renew automatically, so cancel well before renewal; the Terms ask for 30 days. The published refund policy covers only Pro and Grow, 24 hours on monthly plans and 14 days on yearly ones; our pricing guide has the full rules.

For one-off files there is also a free transcription tool on Riverside's website. It takes MP3, WAV, MP4 and MOV files with no sign-up, and Riverside says it handles unlimited files in 100+ languages ^[3]. The trade-off is speaker handling: a multi-speaker upload gets all of its text assigned to one speaker. Files are removed 24 hours after the transcript is generated. That suits a single-voice memo or a lecture. Interviews with guests on separate tracks come out far cleaner from the studio itself, though the Privacy Policy lets Riverside train its AI on them unless a Business contract opts out (our review).

Translating into 30 languages, and what it costs

AI Translation turns a whole video into one of 30 languages, keeping each speaker's voice, with optional lip sync. It is on Pro, Grow, Webinar and Business, even though an older product page calls it Business only. The limits are tight. The recording must include video, run 60 minutes or less, have no more than four speakers and contain no screen share ^[4]. Riverside says processing takes roughly 20 minutes per minute of video, so a 30-minute episode may take up to 10 hours. It translates everything, and VideoDub can switch chosen words back afterward.

Translation costs 100 AI credits a minute. Paid plans include a starting balance of AI credits, but credits do not renew monthly, and extra packs are one-time purchases that work out to $0.02 a credit, or about $2 per minute of translated video. A 30-minute episode needs 3,000 credits, which is exactly the $60 pack. Unused credits are never refunded or carried forward, so buy the pack size you will actually use. Transcription itself does not appear on Riverside's credit cost list, which names translation, AI Twin, voice clones and B-roll.

The bottom line

Riverside's transcription is strongest where its recorder is strongest: guests on separate tracks get named, editable text that feeds captions, clips and edits without extra work. Set the studio language first, proofread names, seed the Dictionary if you are on Grow or above, and never regenerate a corrected transcript. The real decision is the export. Free is not a plan for transcripts, Pro lets you take TXT and SRT files anywhere, and translation adds a per-minute credit cost on top. For most teams that publish transcripts or subtitles, Pro is the plan to start on.