Turn a Thai recording into text you can read and search

One recording, heard and written#

Press play. This is a real recording and the model's own transcript of it, captured and frozen, with each word lighting as it is spoken.

The same recording, written for reading

สวัสดีค่ะคุณสมชายยอดชำระของท่านจำนวน 1,250 บาทครบกำหนดวันที่ 15 สิงหาคมนี้ชำระผ่านแอปได้ตลอด 24 ชั่วโมงค่ะ

Figures were said aloud in the recording and printed here as digits. Both versions come back from the same request, and the speech-to-text page shows them side by side.

paxa-stt-lite-v1-preview · September 9, 2026

What you can send#

The format is read from the file itself. Nothing has to be declared, converted, or renamed first.

• MP3, WAV, FLAC, OGG, M4A, AAC, WEBM

  Common audio and video audio

  The track a phone, a meeting recorder, or a video editor already wrote. An unreadable file is refused before anything is charged.

• 60 minutes

  Per recording

  Longer material is split at a natural break and sent as parts, with each part's seconds offset by where it started.

• 25 MiB

  Per file

  Speech compresses well. An hour of ordinary voice recording sits inside this comfortably at normal settings.

• 9 minutes

  With speaker turns on

  Marking who spoke when reads the whole recording at once, and that work carries its own shorter ceiling.

Thai and English, including a Thai sentence that names a term in English. The transcript keeps both.

What people transcribe#

The same request behind each of these, with the options that suit the material.

• Meetings

  The recording a team already keeps, turned into minutes somebody can read on the way home.

• Call recordings

  Support and sales calls, with the two speakers separated and every line searchable.

• Interviews

  Research and journalism, where a quote has to be exact and has to be findable again.

• Voice messages

  The notes a team sends each other all day, read in a few seconds each.

• Lectures and classes

  A recorded class with a chapter index. A student jumps to the minute a term was defined.

• Video captions

  Subtitle files built from word timings, which is the only way Thai cues break in the right places.

Questions#

Is it free?

Signing in comes with 100 credits and no card, which covers about 12 minutes of audio. That is enough to run a real recording of your own and see the result before deciding anything.

Do I have to write code?

No. Sign in, open the transcribe tool, and upload the file. The transcript comes back in the browser with its word timings, and it can be copied out as text.

What languages does it read?

Thai and English, including a Thai sentence that switches into English for a term or a brand. Both scripts stay in the line, and the words on either side keep their seconds.

Can it tell speakers apart?

Yes, on recordings up to 9 minutes. The transcript comes back as turns numbered by who spoke first, with no claim about anybody's identity.

Are numbers written as digits or as words?

You choose. One convention writes figures the way they were said, and the other writes them as digits, units, and the repetition mark. A search index and a subtitle file want opposite answers.

What happens to my recording?

It is processed to produce the transcript. The privacy policy states how model inputs and outputs are handled, and it is the binding description.

Can I send a video file?

Send its audio track. Every editor and most command-line tools export one in a format on the list above, and the transcript's seconds still line up with the video.

How accurate is it?

No figure goes out without the method beside it. Running your own recording through the free grant answers the question for your material, which is the answer that matters.