Sign in with Google
No separate sign-up; free minutes every month.
120 free minutes every month
Taiwanese Hokkien recordings to text
Upload Taiwanese Hokkien or mixed Hokkien–Mandarin audio. Azhu separates the speakers and writes a Mandarin transcript.
Taiwanese Hokkien recordings, transcribed into readable Mandarin
Elders' stories, field interviews, community meetings: upload Taiwanese Hokkien or mixed Hokkien–Mandarin audio or video. Azhu separates the speakers and writes the transcript in Mandarin, with a summary if you want one.
What does Hokkien turn into?
Hokkien is written out as Mandarin text, not romanization and not Hokkien characters, so anyone can read it, search it, quote it and turn it into a report.
The transcript gives the same meaning in Mandarin rather than a sound-by-sound record. If your work needs the original Hokkien wording or pronunciation, use the transcript as a draft and an index, and go back to the audio for the key passages.
Mixed Hokkien and Mandarin
Choose “Taiwanese Hokkien” when you upload. Use it when some people speak Hokkien and others Mandarin, or when one person switches back and forth; the output is still Mandarin text.
“Auto-detect” is meant for Mandarin, English and other languages. For recordings with Hokkien, choose “Taiwanese Hokkien” yourself.
Good for
- Elders' memories and family history
- Fieldwork and oral history interviews
- Community and neighborhood meetings, clubs and temple committees
- Hokkien talks, programs and videos, with an SRT subtitle file to export
Where mistakes happen
Names of people and places, special terms, strong accents and uncommon expressions are the usual mistakes. List the attendees and terms before uploading and the transcript and summary follow your spelling.
Background noise, people talking over each other and distant microphones also cause errors. After transcription, click any line to listen and fix it.
A transcript in 3 steps
Upload a recording or video
From a computer or phone. Close the page once it's up.
Read the transcript
An hour of audio usually takes minutes. Edit as you listen.
- MP3
- M4A
- AAC
- WAV
- FLAC
- OGG
- OPUS
- WMA
- AMR
- MP4
- MOV
- MKV
- WEBM
- AVI
Can I get romanization or Hokkien characters?
Not at the moment. Hokkien is only written out as Mandarin text.
What about Hakka?
Hakka isn't supported yet.
How accurate is it?
With a clear recording and people taking turns, most of it comes out right; accents, names and overlapping speech are where errors happen. Try one of your own recordings with the free minutes first. Every account gets free transcription minutes each month.
Will my recordings be used to train models?
Speech recognition and AI summaries run on our own computers in Taiwan. Recordings and transcripts are never sent to outside AI services or used to train models.