TL;DR / Key Takeaways
- All four tools scored about 13% WER on the first 5.5 minutes.
- Over 10 minutes: HappyScribe 19.1%, Otter 19.5%, HyNote 21.1%.
- Transkriptor's free plan cut our file at 5:35.
- Otter's free plan allows only 3 file imports per account.
- HyNote and Transkriptor made the same errors word for word.
Short answer: on a messy 10-minute, four-person meeting, HappyScribe, Otter, HyNote and Transkriptor were within about two points of each other: roughly 13% of words wrong over the first five and a half minutes and 19% to 21% over the full ten. Accuracy will not decide this for you. The free plans will: HappyScribe gives you 10 minutes once, Otter caps file uploads at three per account, and Transkriptor stopped our file at 5:35. Tested October 31, 2026.
How we tested: on October 31, 2026 we took the first 10 minutes of a real recorded meeting from the AMI Meeting Corpus (meeting ES2004a, University of Edinburgh, licensed CC BY 4.0): four people with British and European accents, often talking over each other. We uploaded the same MP3 to each tool on a new free account with default settings and English selected. We then scored each transcript against the corpus's human transcript using word error rate (WER): the share of words a tool got wrong, dropped or added. Before scoring we lowercased both sides, removed punctuation and filler words such as "um", and spelled out numbers, so "25" and "twenty five" count as the same. Lower is better. This is one recording, so treat differences of a point or two as a tie.
The scores
| Tool | WER, first 5.5 min | WER, full 10 min | Speakers | Free plan |
|---|---|---|---|---|
| HappyScribe | 12.8% | 19.1% | 4 labelled | 10 minutes in total, once |
| Otter | 13.0% | 19.5% | 4 labelled | 300 min a month, but 3 file imports per account, ever |
| HyNote | 13.4% | 21.1% | Not labelled | Full 10 minutes transcribed |
| Transkriptor | 12.8% | Not available | 4 labelled | Our file stopped at 5:35 |
Two windows, because Transkriptor's free plan only returned the first 5 minutes 35 seconds of our file. Over that stretch all four tools land between 12.8% and 13.4%, which on one recording is a tie. Over the full ten minutes, HappyScribe (19.1%) and Otter (19.5%) edge out HyNote (21.1%). Most of the extra errors in the second half come from crosstalk, while the team took turns drawing animals on a whiteboard and talked over each other. Every tool dropped words there, HyNote a few more than the others.
Where they differed
- Hearing the meeting jargon: the project manager says the team will do "tool training". Only Otter and HappyScribe heard "tool"; HyNote and Transkriptor both wrote "do two training".
- Speakers: Otter, HappyScribe and Transkriptor each found four speakers. Otter's talk-time split (43%, 30%, 16%, 11%) came close to the real one (40%, 27%, 19%, 15%). HyNote's free plan gave one block of text with no speakers.
- Numbers: a currency exchange at the end tripped everyone. Otter merged "seventeen, seventeen pounds" into "1717 pounds"; HappyScribe turned "one point four" into "€1.4".
HyNote and Transkriptor wrote almost the same words
Over the first five minutes, HyNote and Transkriptor produced near-identical text, including the same mistakes: "She going to answer me?", "do two training" and "It doesn't let you put." Two products making the same odd errors usually means the same speech-recognition engine underneath. Neither says which engine it uses, so we cannot confirm it, but if you are choosing between those two, compare the plans and features, not the accuracy.
The free plans, side by side
Otter's free Basic plan includes 300 minutes a month and up to 30 minutes per conversation, but only three audio or video file imports per account, ever, according to its pricing page. That is plenty for live meetings recorded in Otter and very little if you mostly upload files. Pro is $16.99 a month, or $8.33 a month billed yearly, with 1,200 minutes and 10 imports a month.
Transkriptor transcribed in under a minute, but the editor showed our file as 5:35 long and put the rest behind "Unlock Full Transcription". Its AI summary panel showed a sample "Q1 Sales Meeting Summary" that had nothing to do with our meeting, behind an upgrade button. Its pricing page lists a free plan of 90 minutes; on our account, the first file was cut at about five and a half. Pro is $19.99 a month, or $8.33 a month billed yearly, with 2,400 minutes.
HappyScribe is the most upfront: the order screen said our file would use all 10 of our 10 free minutes. That is the whole free plan, once. Basic is $17 a month, or $8.50 billed yearly, for 120 minutes.
HyNote transcribed the full 10 minutes on its free plan and gave the note a sensible title, "Remote Control Project Kick-off Meeting". Speaker labels were missing. Its paid plans are billed yearly: Pro $79.99 a year, Plus $129.99 and Unlimited $189.99.
Enjoying this? Get one like it in your inbox each morning.
one email a day · unsubscribe in two clicks · no third-party tracking
Which one should you use
- You record meetings live: Otter. Speaker labels and keywords were the best of the four, and 300 free minutes a month is generous. Just do not rely on it for uploads.
- You have one file to transcribe: HappyScribe or HyNote. Both returned the full 10 minutes free; HappyScribe adds speakers, HyNote adds summary templates.
- You transcribe a lot of files: compare paid minutes per dollar. Transkriptor Pro (2,400 minutes for $8.33 a month billed yearly) is the most minutes for the money of these plans.
Related reading
Frequently asked questions
Which AI transcription tool is most accurate?
In our October 31, 2026 test on a real four-person meeting, HappyScribe, Otter, HyNote and Transkriptor were within about two points: 12.8% to 13.4% word error rate on the first 5.5 minutes, and 19.1% to 21.1% on the full 10 minutes for the three that transcribed it all.
Is Otter free for uploading recordings?
Otter's free Basic plan allows only three audio or video file imports per account, ever, although it includes 300 minutes a month of transcription for live meetings.
Why do transcripts get worse in meetings?
Crosstalk. When several people speak at once, every tool we tested dropped words. Our error rates rose from about 13% in the calmer first half to 19% to 21% over the full meeting.