Riverside

The transcription page accepts MP3, WAV, MP4 and MOV files without an account. You pick the language and content type, let the AI run, and download the result as a plain TXT file or a timestamped SRT subtitle file. Riverside says the tool can be used as much as you want, and it describes accuracy of up to 99% across more than 100 languages; those figures are the provider claims, not a measured result.

The limit that matters is speakers. The page FAQ says that when you upload multi-speaker audio or video, all the text is assigned to one speaker. That makes the free tool fine for captions or a searchable text of a solo recording, but an interview transcript with named speakers needs recordings made inside Riverside on separate tracks.

The official trial guide, updated September 10, 2026, states that Riverside no longer offers a free plan and replaced it with a 14-day free trial of Pro, Grow or Webinar. The help center adds that a credit card is needed to start, and the card is charged automatically on day 15 unless the trial is cancelled in Settings under Plan and Billing.

Many reviews still describe a free account with 2 hours of multi-track recording and a watermark. That block remains hidden in the pricing page code, and the help article on free accounts now asks for a sign-in. Plan around the trial: cancelling keeps existing recordings in the account, but further editing needs a paid plan.

Recordings made in the Riverside studio are transcribed automatically after the session ends, usually within minutes depending on length. Speaker detection relies on separate audio tracks: each participant recorded on their own track gets a label, while several people on one microphone, or an uploaded mixed file, are not separated.

Set the recording language in the studio settings before the session. In the editor, the transcript is synced with the timeline, so deleting words cuts the video, Correct everywhere fixes a repeated misspelling, and captions are generated from the same text. Transcripts download from the Tracks section as a text file or as subtitles.

All paid plans include unlimited recording and transcription; they differ in studios, separate track download hours and live features. Editorial assessment: a single show recorded remotely fits Pro, while Grow is worth it only for live streaming or a second show, and Webinar only when registration, polls and lead capture are needed.

• Pro: $29 a month, or $24 a month billed annually; 1 studio; 15 hours of separate track downloads a month; 2,000 starting AI credits.
• Grow: $39, or $34 annually; 2 studios; 20 hours; full HD live streaming and multistreaming.
• Webinar: $99, or $79 annually; 3 studios; 25 hours; webinars with up to 100 registrants.
• Business: custom quote; unlimited studios and downloads, SSO, API access and collaborative editing.

The refund policy allows a full refund, on request, for a new monthly Pro or Grow subscription cancelled within 24 hours of purchase, or a yearly one cancelled within 14 days. AI credit purchases are not refunded or carried forward, and subscriptions bought through Apple or Google follow the store policy, which the terms describe as final sale.

A downgrade keeps the current benefits until the billing period ends, and the help center recommends downloading separate tracks and exported timelines before then. The terms require 30 days of notice before a price change, with a prorated refund of prepaid fees if you cancel before the new price applies.

The terms let Riverside use inputs and outputs of its AI tools to train, fine-tune and improve its own models, while third-party AI providers may not train on them. The privacy policy counts transcripts and content derived from recordings as Content Data that can be used for training, except for Business customers who opt out under a written agreement.

The same policy makes the account holder responsible for giving notice to and obtaining consent from everyone whose voice or image is recorded. Features that build an AI voice or face from your samples, such as AI Twin, process that data only with your consent and keep it only as long as the feature requires. The terms also warn that AI output may not be unique.

1. Select English and the matching content type, then start the transcription.
2. Download both the TXT and the SRT files.
3. Load the SRT into a video player and check caption timing at the start, middle and end.
4. Count how many of the 10 listed terms are spelled correctly.
5. Note whether the TXT file separates the two speakers.

Accept captions that stay in sync and a term list with no more than two errors. Expect a single speaker label, as the FAQ states, and reject missing sections or captions that drift out of sync. No accuracy figure is reported until the exercise is run.

To edit a recording by its transcript without Riverside recording, compare Descript. For live noise removal and bot-free meeting notes, see Krisp, and for cleaning a finished voice track, Adobe Podcast.