Comparison
txscribe vs Sonix, for interviews and research
Sonix and txscribe want the same thing from a transcript: that you can check it. Both link the transcript to the audio, and both say they don't train on your recordings. Sonix also marks the words its recogniser was unsure of, which we don't, and it connects to research software.
Where the two differ most for one person transcribing interviews is in what an hour costs and how many languages are covered — the table sets it out.
Sonix's details checked on its own pages on ; each links to its source. Plans and prices change — if something here is out of date, tell us at support@txscribe.ai.
Price
txscribe
Pro: $24.99 a month, billed monthly. One plan, no seats.
Sonix
Pay As You Go: $10 an hour, no subscription. Core $25 a month or $275 a year; Advanced $50 a month or $550 a year; Pro $80 a month or $880 a year — each with 1 user included, extra seats $25 a month or $275 a year. Source
Time included
txscribe
Pro: 30 hours a month. Without an account: one recording, up to 10 minutes.
Longest recording
txscribe
Up to 8 hours per file on Pro. Without an account, the first 10 minutes of one recording.
Languages
txscribe
82 languages (110 regional variants in the picker). Speaker labels offered in all of them — in 96, only on recordings of up to 2 hours, and not where that pass fails.
Speaker names
txscribe
Speakers come back numbered. Rename one once and every line follows; names you have used before are offered again.
Exports
txscribe
TXT, DOCX, PDF, SRT, VTT, Final Cut Pro captions, Premiere Pro markers, transcripts for NVivo, MAXQDA and ATLAS.ti, and JSON, on every plan, including Free. Up to 100 transcripts at once as one ZIP, and the audio or video file you uploaded can be downloaded again.
Translation
txscribe
A translated copy of the whole transcript, on every plan.
Custom word list
txscribe
Up to 1,000 names and terms, on every plan.
Checking against the audio
txscribe
Words come back with timestamps: click any line to jump to that moment in the recording. A line you correct keeps its paragraph's timing rather than per-word timing. Answers to questions about a transcript link to the moments they came from.
Human transcription
txscribe
No human transcription. The machine's draft is yours to correct, in an editor built for it.
Sonix
Human transcription isn't sold by Sonix itself: it lists independent transcriptionists you can hire, and says it doesn't employ them directly. Source
Training on your recordings
txscribe
We do not use your recordings or transcripts to train models.
Where Sonix is the better choice
- It shows you where to look. Click any word to jump the audio there; a heatmap shades each word by the recogniser's confidence, so uncertain words stand out. Sources 1, 2 txscribe doesn't mark uncertain words; you find errors by listening.
- Recordings that switch language. 54 languages on every plan; a recording in more than one language can be transcribed as such, up to 2 hours 15 minutes long. Sources 1, 2, 3 In txscribe, speech in a second language may come back translated instead of transcribed, or not at all.
- Research software. Exports to the research tools NVivo, ATLAS.ti and MAXQDA that keep speaker labels, timestamps and coding markers. Source Our Word and text exports carry speakers and timestamps, but nothing made for those tools.
- Questions across a whole project. On its subscription plans (not Pay As You Go), AI Workspaces answer questions across several transcripts, with citations that link to the recording and timestamp. Sources 1, 2 txscribe answers questions across your whole library or one folder too, but it reads at most four recordings per question: the ones that use the question's own words.
- Timing that follows your edits. After you edit a transcript, one click realigns every word's timing to the audio. Source A line you correct in txscribe keeps its paragraph's timing rather than per-word timing.
- A security report to hand over. Its security page says SOC 2 Type II certified; its pricing gives access to the report on Pro. Sources 1, 2 We have no SOC 2 report.
What txscribe doesn't do well
- Speaker labels stop at 2 hours in 96 of our 110 language variants: there a separate pass labels the speakers, and a longer recording — or one whose pass fails — comes back without labels and says so, with no speakers to rename.
- A recording that switches language partway may come back translated rather than transcribed — or not at all — and nothing marks it as missing. Name the main language instead of leaving it on automatic.
- Summaries do not point back to the moment each line came from. Answers to questions do.
Try it on one of your recordings.
Up to 10 minutes, no account and no card. A table can tell you the limits; only your own audio tells you whether the transcript holds.
Start free