Comparison
We compared two recordings processed by both tools. Below are the method, observed differences and limitations of the test.
Two real recordings: a 48-minute Estonian discovery call and a 27-minute English client call with a three-minute Estonian side-conversation in the middle. Each was processed twice, once by each tool. Participants and their commercial details are anonymised throughout.
We did not have an independent reference transcript. Our output was used for alignment, so this is not a neutral accuracy benchmark. The examples describe the two recordings tested; they do not establish overall vendor accuracy.
On the English call, three people were in the room. MeetGeek’s export contains six speaker identities, including one assigned words that were not spoken. Roughly one turn in nine is attributed to “Unknown speaker”. These counts come from its export. Teneks returned three named speakers.
On the Estonian recording, MeetGeek returned three labels. Teneks returned three labels and left six short replies unattributed.
About one word in twenty-six of MeetGeek’s Estonian transcript falls within two repetition runs: one of 106 tokens and another of 84. Both appear near the end of the recording. Teneks’s longest run was five repetitions of a word the speaker said five times.
The repetition appears in the export for this recording. We have not established its underlying cause.
Two participants spoke Estonian for about three minutes while waiting for a screen share. MeetGeek returned English sentences that were not spoken and assigned them to a speaker who was not present. One Estonian word became “What’s up? What’s up? What’s up?”
Check the cited audio alongside the export when evaluating this result.
Two recordings cannot establish the cause of a transcription error or predict how a future model will perform. Audio conditions, configuration and product versions may affect the result. Repeat the test with recordings representative of your team.
MeetGeek’s English transcript was smoother to read because it removed disfluencies. Its repetition behaviour on that recording was also slightly better than Teneks’s. Teneks kept more of the original wording. Review the format that suits your team’s follow-up and coaching work.
Include your usual languages, shared microphones and audio conditions in a trial to see whether the differences matter for your team.
Teneks also produced errors, including duplication and attribution issues. Both tools mis-transcribed a prominent company name on the English call. Review these limitations alongside the differences above.
Compare both tools on a recording with your usual languages, speakers and audio conditions. We can help you review the results.