Transcribe Thai audio to text, free

What makes Thai hard to transcribe

Thai is written with no spaces between words — spaces mark clause breaks instead — so word boundaries have to be inferred, and accuracy is measured per character. Thai also has five tones marked by a combination of tone marks and consonant class, so the written form of a tone is not a single symbol you can simply attach.

What your Thai transcript looks like

Text comes back as continuous Thai script with spaces only where a Thai writer would put them, at phrase boundaries.

[00:00 → 00:04] สวัสดีครับ วันนี้เราจะเริ่มประชุมกันนะครับ

[00:04 → 00:08] ก่อนอื่นผมขอพูดถึงผลงานของเดือนที่แล้ว

An illustration of how Thai transcripts are formatted — script, punctuation and timestamps. Not the output of any particular recording.

How accurate to expect

News and presentation Thai transcribes solidly. Fast conversational Thai with heavy particle use is the hardest case, and word boundary decisions become less reliable there.

What usually needs fixing

1. 1 Word segmentation, since Thai gives no visual clue where one word ends and the next begins.
2. 2 Proper nouns, especially Thai personal names, which are long and often unfamiliar to the model.
3. 3 English technical terms mixed into Thai speech, which may appear transliterated into Thai script rather than left in Latin.

Getting better results

• → Thai speech often mixes in English technical terms; these appear in Latin script inside the Thai text.
• → Formal news-style Thai transcribes noticeably better than fast conversational speech.

Who transcribes Thai