What is the difference between verbatim and clean?

Verbatim writes every word as spoken, false starts included. Clean removes those repairs to make the note readable, which makes it a looser record of what was said. A file that has to hold the exact words wants verbatim.

Will it spell our part numbers correctly?

Send them as vocabulary, up to 50 terms. Part numbers, site names, and trade words then carry one spelling across every note the team dictates.

Can somebody dictate in Thai and English together?

Yes. A Thai note that names a product or a standard in English keeps the English, and the Thai around it stays Thai.

Is this a record-keeping system?

It returns the text. Where that text is filed, how long it is kept, and whether it satisfies a professional obligation are decisions your own system makes.

All use cases

Thai speech to text for dictation

Speak a note and file it as a record. Thai dictation with a verbatim or a tidied transcript, your own terminology, and figures a report can print.

Try it now Read the docs

What the transcript says

A site inspection note

spokenตรวจอาคาร บี ชั้น สาม พบรอยร้าวผนังยาวประมาณ หนึ่งจุดสอง เมตร แนะนำให้ซ่อมภายใน สิบสี่ วัน

writtenตรวจอาคาร B ชั้น 3 พบรอยร้าวผนังยาวประมาณ 1.2 เมตร แนะนำให้ซ่อมภายใน 14 วัน

Measurements and a deadline in one spoken sentence. The written convention prints them the way the inspection report prints them.

A loss adjustment note

spokenเคลมเลขที่ เอ วาย หก แปด สอง ประเมินความเสียหายเบื้องต้น สี่หมื่นห้าพัน บาท รอเอกสารจากอู่

writtenเคลมเลขที่ AY682 ประเมินความเสียหายเบื้องต้น 45,000 บาท รอเอกสารจากอู่

A reference number spoken letter by letter, and an amount. Both land in the fields a claim system expects.

An attendance note

spokenประชุมกับลูกค้าเมื่อ บ่ายสองโมง ตกลงส่งร่างสัญญาภายในวันศุกร์ ค่าบริการตามที่เสนอไว้เดิม

writtenประชุมกับลูกค้าเมื่อ 14:00 ตกลงส่งร่างสัญญาภายในวันศุกร์ ค่าบริการตามที่เสนอไว้เดิม

A record made the same afternoon. The time is written as a clock reading, which is how the file will be read later.

Every transcript pair on this page is an authored illustration of what the convention option changes, re-checked against live output before publish. The Speech-to-Text page carries the model's own witnessed transcripts of a real recording, with the capture date.

How it fits a working day

An inspector finishes a site walk and speaks two minutes into a phone. A surveyor describes damage while standing in front of it. A lawyer records what was said in a meeting before the detail goes. All three are dictation: somebody speaking on purpose, to a machine, to leave a document behind. The value is in the gap between speaking and typing. A note spoken in ninety seconds takes ten minutes to type up at a desk that evening, and half of them never get typed. Transcription closes the gap on the same day, in the speaker's own words, with the terminology of the trade spelled the way the trade spells it.

Record the note on the device already in handA phone recording is enough. One note is one request, and a request takes up to 60 minutes.
Choose the style and the conventionClean for a note somebody will read, verbatim for a record that has to hold every word. Written figures for anything with a measurement or a reference in it.
File the transcript against the recordThe response carries the text and the credits charged. Your system files it under the job, the claim, or the matter it belongs to.

A spoken note, filed

note-4471.txt

ref      4471
author   ผู้ตรวจสอบอาคาร (staff 218)
spoken   2026-09-10 16:42  ·  1:30
style    clean

ตรวจอาคาร B ชั้น 3 พบรอยร้าวผนังยาวประมาณ 1.2 เมตร
บริเวณเสาริมทางเดินฝั่งตะวันออก ยังไม่พบน้ำรั่วซึม
แนะนำให้ซ่อมภายใน 14 วัน และนัดตรวจซ้ำวันที่ 25 ก.ย.

The header is your system's. The body is the transcript, filed under the record it belongs to. The figure shows the shape of a filed note.

A working day, four notes

09:20After the first visitNinety seconds spoken in the stairwell while the detail is fresh. The file uploads before the next appointment starts.

12:05In the carTwo visits described back to back. Both come back as text before lunch is over, each under its own reference.

15:40In front of the damageA measurement and a claim number, said aloud. The written convention puts the figure and the reference in the form the claim system reads.

17:30The day is already writtenThe notes that used to wait for an evening at a desk are text, in the speaker's own words, filed the same day.

A note of 90 seconds is one request. Nothing has to be batched, and nothing waits for a queue.

The three settings dictation uses

verbatim | clean
How closely to follow the speaker: Verbatim keeps every word as spoken. Clean drops the false starts and repairs for readability, which makes it the less exact of the two.; written
Figures as a report prints them: Measurements, amounts, dates, and clock times arrive as digits and units, ready for the fields they belong in.; 50
The terminology of the trade: Part numbers, site names, and the words a profession uses go in as vocabulary and keep one spelling across a team.

Every setting combines with every other, and none of them changes the charge.

What a team of dictated notes costs

Workload	Audio	Credits
One note, 90 seconds	90 s	12.5
One person, one day, 20 notes	30 min	250
One person, a month	11 h	5,500
A team of 8, one month		44,000 credits

8.33 credits a minute of audio, billed by the millisecond. 88 hours of dictation over 22 working days at that rate. Style, vocabulary, and convention cost nothing extra.

For developers

The snippets send one dictated note with the written convention and the team's terminology, then read the transcript back.

AUDIO=$(base64 < meeting.m4a | tr -d '\n')
curl -X POST https://api.paxalabs.com/v1/stt \
  --max-time 300 \
  -H "Authorization: Bearer $PAXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"audio\": \"$AUDIO\", \"model\": \"paxa-stt-lite-v1-preview\", \"convention\": \"written\"}"

AUDIO=$(base64 < meeting.m4a | tr -d '\n')
curl -X POST https://api.paxalabs.com/v1/stt \
  --max-time 300 \
  -H "Authorization: Bearer $PAXA_API_KEY" \
  -H "Content-Type: application/json" \
  -d "{\"audio\": \"$AUDIO\", \"model\": \"paxa-stt-lite-v1-preview\", \"language\": \"th\", \"vocabulary\": [\"Paxa Labs\", \"quarterly review\"]}"

import { readFile } from "node:fs/promises";

const audio = (await readFile("meeting.m4a")).toString("base64");

const response = await fetch("https://api.paxalabs.com/v1/stt", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.PAXA_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({ audio, model: "paxa-stt-lite-v1-preview" }),
  // A long recording can run for minutes; give it room.
  signal: AbortSignal.timeout(300_000),
});

if (!response.ok) throw new Error(`Transcription failed: ${response.status}`);
const { text, usage } = await response.json();
console.log(text, `${usage.credits} credits`);

Full API reference

What the model was built to hear

Paxa STT Lite transcribes Thai and English, including a Thai note that says a part number or a site name in English. It publishes no accuracy figure until the evaluation method is published beside it.

Common questions

What is the difference between verbatim and clean?: Verbatim writes every word as spoken, false starts included. Clean removes those repairs to make the note readable, which makes it a looser record of what was said. A file that has to hold the exact words wants verbatim.
Will it spell our part numbers correctly?: Send them as vocabulary, up to 50 terms. Part numbers, site names, and trade words then carry one spelling across every note the team dictates.
Can somebody dictate in Thai and English together?: Yes. A Thai note that names a product or a standard in English keeps the English, and the Thai around it stays Thai.
Is this a record-keeping system?: It returns the text. Where that text is filed, how long it is kept, and whether it satisfies a professional obligation are decisions your own system makes.

Start with free credits

Start free

More use cases