We ran a 34-minute deposition through AutoScriptOne's transcription and through Stenobox, then scored both against the reporter's certified transcript. Same audio, same scoring. AutoScript came back with twice the errors, every speaker as "Speaker A, B, C," and the witness's mother, sister and insurer misspelled.
AutoScript's text is the file it returned in October 2026, as downloaded, on its pay-as-you-go upload service at $15 an audio hour. Stenobox's is the transcript as delivered, with the case caption attached the way a reporter attaches it to a job. Every quote on this page is copied from one of those two files, and both were scored, unedited, against the certified transcript with the same open method as our MAXScribe comparison.
On the left, AutoScript's file as downloaded; on the right, Stenobox's transcript as delivered. The highlighted words are the ones a reporter fixes by hand, against the audio.
Word accuracy against the reporter's certified transcript of this deposition, 4,597 words, insertions included, every tool scored the same way.
| Tool | Right words | Errors | Structure |
|---|---|---|---|
| Stenobox | 99.28% | 83 | 208 Q. and A. pairs, BY MS. WODIKA, speakers named from the caption |
| AutoScriptOne | 97.41% | 163 | None — 397 turns labeled Speaker A, B, C |
| Otter | 96.04% | 182 | None |
| MAXScribe AutoTranscript | 94.82% | 238 | None — SPEAKER 2, 3, 4 |
"Right words" counts wrong and missing words against the certified transcript; "errors" adds the extra words a tool put in. Stenobox's 83 include eleven name errors of its own on this deposition, counted the same as everyone else's. Otter and MAXScribe were run in July 2026 on the same audio; AutoScript and this Stenobox run in October 2026.
Counted across the whole file. None of it is marked; you find it by listening to all of it.
"Speaker A, B, C" turns to turn into Q., A. and names
Q. and A. lines, BY headers or named speakers
answers with the next question's "Okay" glued on
names and proper nouns wrong
numbers wrong
words missing altogether
| To make it certifiable, you must… | AutoScriptOne | Stenobox |
|---|---|---|
| Know who is speaking | Speaker A, B, C — never a name, and the witness lands under C once | BY MS. WODIKA, Q. and A., THE WITNESS, MR. MILLENBACH |
| Keep each turn whole | The next question's "Okay" glued onto 23 answers | One turn, one label |
| Spell the names and terms | Fixed by hand, every one | Follows the caption you attach |
| Put it on the court's page | Running text: no caption, line numbers, page numbers or certificate | 25 numbered lines, labels at the indent, caption, index and certificate |
| Know where to listen | All of it | The flagged spots, each a click from the audio |
| Your actual job | Rebuild, then verify | Just verify |
Everything a reporter needs to finish the transcript is in the plan, not a separate purchase.
Seven days free. Upload the audio with your caption and compare what comes back to your own transcript.
Try Stenobox free →How we compared: AutoScriptOne's upload transcription, October 2026, pay-as-you-go at $15 an audio hour ($8.47 for this 33-minute file), the file as downloaded. Stenobox, October 2026, the transcript as delivered with the case caption attached, unedited. Both scored against the reporter's certified transcript with the same open method as our MAXScribe comparison; "…" marks a cut. AutoScript's monthly plans run $75 to $350 and are built around its own live-capture recorder; the upload service is what we tested. Its pricing, plan by plan → Prices as published that month; check the current ones. AutoScriptOne and VoiceScript are trademarks of their owner; Stenobox is not affiliated with or endorsed by it. Every transcript requires professional review before delivery or certification.