[002] · SPECIMEN FROM ARCHIVE · FIG. 01 · INTERVIEW TRANSCRIPT · NATURE: MULTIVOICE
Speech, transcribed.
Twenty-eight hours for one staff stenographer. Two minutes for our pipeline. Audio in, correctly spelled text out, with a timestamp on every word. Thirty languages. One export.
[001A] · GUEST SESSION · NO SIGNUP
Drop a file — get a transcript.
Up to 5 minutes per sample. MP3 · WAV · M4A · MP4 · MOV. Auto language detection. No card, no account, no watermark — for this issue only.
or drag here
INPUT
interview-2026.04-MD.mp4
LANG
EN-US (AUTO)
MODEL
grok-stt-v3.1
DURATION
00:34:18
SPK A
[00:00:01.240]
OK. Recording running. Confirm.
SPK B
[00:00:03.880]
Confirmed. For the record — fourteen oh-eight hours.
SPK A
[00:00:08.110]
Noted. Tell me when you first realized the acceleration in the current system was non-linear.
SPK B
[00:00:14.620]
It happened on the second week of calibration. We attributed the drift to thermal noise at first, but —
SPK B
[00:00:21.040]
— by the third iteration it was clear the anomaly had a phase structure.
FIG. 01 · INTERVIEW TRANSCRIPT · NATURE: MULTIVOICE
REC ●
Datasheet of the instrument.
⊐ ⊏
LANGUAGES
30+
recognition · auto-detect
⊕
SPEED
< 02s / min
cloud pipeline · streaming
◉
ACCURACY
96.4%
WER-clean · LibriSpeech reference
▮▯▮▯
EXPORTS
× 9
SRT · VTT · TXT · JSON · PDF · DOCX · SBV · ASS · CSV
Four operations from file to transcript.
01
INTAKE
Upload the specimen
Drag your file in or paste a URL. We accept MP4, MOV, MKV, MP3, WAV, FLAC, M4A, OGG, OPUS, AAC. Up to eight hours per pass.
$tx upload --in=interview.mp4 --lang=auto
02
PARSE
Distill the speech
Audio is extracted, normalized to −23 LUFS, denoised and segmented. Every phoneme is timestamped to the millisecond.
$tx denoise --target=-23 --segment
03
SYNC
Assign the speakers
Diarization labels each turn A/B/C. Names are editable in one click. Overlaps are annotated on a dedicated track.
$tx diarize --speakers=auto
04
ISSUE
Export the artifacts
Nine formats with one command. Defaults: SRT for video, DOCX for journalists, JSON for engineers. Clean, no marks.
$tx export --fmt=srt,vtt,docx,json
Recognized sources.
English
en
Portuguese
pt
Brazilian PT
pt-BR
Spanish
es
French
fr
German
de
Italian
it
Russian
ru
Ukrainian
uk
Polish
pl
Dutch
nl
Swedish
sv
Norwegian
no
Danish
da
Finnish
fi
Turkish
tr
Arabic
ar
Hebrew
he
Hindi
hi
Bengali
bn
Japanese
ja
Korean
ko
Mandarin
zh
Cantonese
zh-yue
Indonesian
id
Malay
ms
Thai
th
Vietnamese
vi
Filipino
fil
Catalan
ca
Greek
el
Hungarian
hu
Romanian
ro
Nine export formats.
| FORMAT | EXT | DOMAIN | SPEC |
|---|---|---|---|
| SRT | .srt | Subtitles — video players | SubRip · UTF-8 · 1-indexed · CR/LF |
| VTT | .vtt | HTML5 web players | WebVTT · style blocks optional |
| TXT | .txt | Clean printout | no timestamps · paragraphs by speaker |
| JSON | .json | Engineers · pipeline | word-level array · ms precision |
| Lawyers · archive | A4 · pdf/A-1b · signature optional | ||
| DOCX | .docx | Journalists · editors | Office Open XML · speaker styles |
| SBV | .sbv | Legacy YouTube · upload | YouTube subtitle format |
| ASS | .ass | Karaoke · stylization | Advanced Substation · layered |
| CSV | .csv | Analysis · spreadsheets | one row per segment · ; separator |
Four tariffs. One counter.
QUOTATION N.º 0001/2026 · A
FREE
Lab trial
€0/ mo
- 014,000 credits · one-time
- 02≈ 130 minutes of transcription
- 03all 9 export formats
- 04«transcript.pt» mark in PDF/DOCX
- 05queue priority: standard
QUOTATION N.º 0001/2026 · B
GROWTH
Solo desk
€27/ mo
€19/mo billed yearly
- 01142,000 credits · monthly
- 02≈ 4,700 minutes of transcription
- 03no watermark
- 04dubbing in 30+ languages
- 05API token · sandbox tier
MOST CHOSEN
QUOTATION N.º 0001/2026 · C
PRO
Team
€49/ mo
€34/mo billed yearly
- 01260,000 credits · monthly
- 02≈ 8,600 minutes of transcription
- 03everything in Growth
- 04API + webhooks · production tier
- 05priority queue · dedicated support
QUOTATION N.º 0001/2026 · D
MASSIVE
Studio
€120/ mo
€84/mo billed yearly
- 01600,000 credits · monthly
- 02≈ 20,000 minutes of transcription
- 03everything in Pro
- 04extended commercial license
- 05dedicated session with engineer
Benchmarks against the field.
| METRIC | TRANSCRIPT.PT | OTTER.AI | REV.COM | DESCRIPT |
|---|---|---|---|---|
| WER (avg) | 3.6% | 6.1% | 4.0% (human) | 5.4% |
| Languages | 30+ | < 7 | ~ 35 (EN AI only) | ~ 23 |
| Max file | 8 h | 4 h | 5 h | 10 h |
| Word timestamps | Yes · ms | Yes · ms | Pro tier only | Yes · ms |
| Spike throughput | 90 min/min | ~ 10 min/min | human — N/A | ~ 30 min/min |
| EUR billing | Yes · with NIF | USD only | USD only | USD only |
| No watermark | Paid tiers | Paid tiers | Yes | Paid tiers |
| Free tier | Yes · 130 min | Yes · 300 min/mo | None | Yes · 60 min/mo |
Ten frequently asked, by incidence.
Q01 / A01
What does transcript.pt actually do?
It converts spoken audio into synchronized written text. A file goes in; out comes a readable transcript with word-level timestamps and speaker labels. No camera, no hands, no copyediting.
Q02 / A02
Which input formats do you accept?
MP4, MOV, MKV, MP3, WAV, FLAC, M4A, OGG, OPUS, AAC. Plus URLs from S3, Vimeo, YouTube, Google Drive. Video is split into tracks; the extracted audio enters the STT pipeline.
Q03 / A03
How long does an hour-long file take?
Twenty to thirty seconds on the standard queue. About ninety seconds under load. Pro plan runs on a priority queue regardless of traffic.
Q04 / A04
Do you support European Portuguese, not only Brazilian?
Yes. pt-PT and pt-BR are separate dialect models. Auto-detect picks by phonetics; force it with --lang=pt-PT.
Q05 / A05
How do you label speakers?
Diarization in Grok STT v3.1. Default labels A/B/C/... Names are editable from the UI in one click; the export is re-saved automatically.
Q06 / A06
Are exports watermarked?
On the free tier, a discreet «transcript.pt» line appears in the footer of PDF and DOCX. All paid tiers export clean.
Q07 / A07
What is your retention policy?
Files stay in your account until you delete them. AWS S3, eu-central-1 region. Stripe EUR billing. Deletion-on-request is final and logged.
Q08 / A08
Can I get dubbing and lip-sync after the transcription?
Yes. One click in the same session — we translate to 30+ languages and pipe to the lip-sync engine. Dubbing is billed by output-video minutes.
Q09 / A09
Is there a developer API?
There is. POST /v1/transcribe for synchronous calls; webhooks for async. Sandbox token on Growth, production tier from Pro upward. Documentation at /docs.
Q10 / A10
Who runs transcript.pt?
SPACEFOX UNIPESSOAL LDA, registered in Fernão Ferro, Portugal. The same team has run doitong.com since 2023. We read hello@transcript.pt and reply in the language we receive.
Five entries from real work.
REPORT N.º 0142
2026-04-12
— Six hours of source recording. Transcribed in eleven minutes, speakers labeled correctly. I tagged the names and went straight into the text. What used to take two nights took a lunch break.
REPORT N.º 0156
2026-04-19
— Parliamentary hearing — four and a half hours. Transcript with a timestamp on every word. Quoted in my thesis to the millisecond. My supervisor asked how — I didn't say.
REPORT N.º 0181
2026-04-23
— I publish episodes weekly. The transcript becomes the show notes; I edit in DOCX, save back — captions re-sync themselves. A six-hour workflow shrunk to forty-five minutes.
REPORT N.º 0204
2026-04-25
— I've been recording lectures on my phone for a year and a half. Uploaded them in series — got a word-searchable archive back. Students prepare faster now. So do I.
REPORT N.º 0218
2026-04-28
— Used for witness depositions. Signed PDF/A-1b drops straight into the case file. Courts accept without challenge. Cost per case is below half an hour of paralegal time.
[011] · CALL
Open the lab.
A starter pack lands in your account on signup. No card. No commitment. The cancel button sits one click away from the dashboard.