Transcribe Vietnamese audio to text, free

What makes Vietnamese hard to transcribe

Vietnamese writes six tones using diacritics stacked on top of vowel marks — the same three letters can be ma, má, mà, mả, mã or mạ, and each is a different word. Most speech engines built for English drop these marks or guess them from context, which turns a transcript into something a Vietnamese reader has to decode rather than read.

What your Vietnamese transcript looks like

Transcripts come back with tone marks in place, so 'cà phê' does not become 'ca phe'. Sentence-final particles like nhé, ạ and à are kept rather than trimmed as noise.

[00:00 → 00:04] Chào mọi người, hôm nay chúng ta bắt đầu buổi họp lúc chín giờ.

[00:04 → 00:09] Trước tiên, tôi muốn nói về kết quả của tháng trước.

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

How accurate to expect

Clear studio or headset recordings transcribe reliably. Accuracy drops fastest on street-noise recordings and on fast overlapping conversation, where tone marks are the first thing to go.

What usually needs fixing

1. 1 Proper nouns — Vietnamese personal and place names are short and tonal, so a misplaced tone mark turns one name into another. Check names before publishing.
2. 2 Numbers spoken quickly can merge, since Vietnamese number words are single syllables with no natural pause between them.
3. 3 Regional vocabulary from the Mekong Delta is less represented in training data than Hanoi or Saigon speech.

Getting better results

• → Northern and Southern accents differ enough that proper nouns can land differently — read names once before publishing.
• → Mixed Vietnamese-English speech is common in tech and business recordings; the model handles the switch without you splitting the file.

Who transcribes Vietnamese