Cantonese and English interviews
Notes you can check.
An interview recording goes in. A timestamped transcript comes out, with a summary, key points and action items where every single point shows the moment it came from.
3 hours of audio free, no card. Upload only — nothing joins your calls.
Summary
Summary in EnglishA 42-minute offer discussion between a recruiter and a candidate. The base salary was set at HK$45,000 , and the sign-on bonus is still waiting on HR .
Key points
- The candidate asked for HK$45,000, not an earlier figure, to go in the letter
- One-month notice period, start date early November
Action items
| Task | Owner | Due | Evidence |
|---|---|---|---|
| Send the revised offer letter | Ms. Chan | 30 Oct 2026 | |
| Confirm the sign-on bonus with HR | not specified | not specified |
Transcript
Cantonese and English · 128 segmentsHow it works
Four steps, and you can close the tab during the slow one.
- 01
Upload a recording
Choose a file of up to 2 hours and 1 GB. MP3, M4A, WAV, FLAC, OGG, AAC, WEBM and more. Nothing joins your call and nothing needs to be installed.
- 02
Read the transcript
Timestamped segments in the language that was actually spoken, with anonymous speaker labels wherever the speech engine separated the voices.
- 03
Check the summary
A summary, key points and action items. Each one carries the timestamps it came from, and an owner or a date appears only if somebody said it aloud.
- 04
Download the notes
TXT, DOCX and JSON, all written from the same stored result, so the file you send a colleague says what the screen said.
Languages
Built for Cantonese and English, including both at once.
A Hong Kong interview rarely stays in one language for a whole sentence, let alone a whole hour. Recolio is set up for that case rather than for one language at a time.
- Language identification
- Azure AI Speech is given Cantonese and English as candidates and picks between them phrase by phrase, so a sentence that starts in one and finishes in the other survives intact.
- Speaker labels
- Speaker 1, Speaker 2, and so on. Anonymous on purpose: Recolio separates voices, it does not work out whose they are.
- Timestamps
- Every segment carries its position in the recording, which is what makes a citation something you can act on rather than read.
- Original-language quotes
- The transcript keeps the words as spoken. The summary can be written in English or Traditional Chinese and is labelled with which, so a translation is never mistaken for a quote.
Grounding
A summary is a claim. This one comes with the evidence.
Every key point and every action item carries the timestamps of the segments it was drawn from. Clicking one scrolls the transcript to that moment and flashes it, so the distance between reading a claim and checking it is one click rather than a replay.
What the summary says
The sign-on bonus was not committed to and is waiting on HR 12:31
What was said at 12:31
Speaker 1:Sign-on 要再同 HR confirm，我唔想而家俾一個數你。Low confidence
What we do not claim
Recolio does not promise a correct transcript, and it does not publish an accuracy figure it has not measured on your kind of audio. Where the speech engine reports low confidence in a segment, you get a badge rather than a probability, because a model's confidence score is not a measurement and should not be dressed up as one. Where a summary has no owner or no date to report, the field reads not specified instead of a plausible guess.
Pricing
Introductory pricing, one seat at a time.
These prices are introductory while we work with our first teams, and they may change before general release. Anyone already subscribed will hear from us before that happens.
Per seat
IntroductoryUS$20per seat, per month
- 10 hours of audio per seat, pooled across the whole workspace
- Extra hours at US$3 each, prepaid, and they never expire
- Invite teammates; each person sees only their own recordings
- TXT, DOCX and JSON downloads on every result
3 hours of audio, no card, once per workspace.
- How hours are counted
- An hour of allowance is an hour of audio, measured on the decoded recording. You are charged once when the transcript is ready, and not at all if transcription fails.
- Extra hours
- US$3 per hour, prepaid. Prepaid hours never expire and are only used once the month's seat hours are gone.
- At renewal
- The month's unused seat hours do not carry over; the new month replaces them. Prepaid hours are untouched by a renewal.
- Who this is for
- Research, recruitment and consulting teams who work through interview recordings, in Hong Kong first.
The plain facts
Everything worth knowing before you upload anything.
- Formats
- MP3, M4A, WAV, FLAC, OGG, AAC, WEBM and more.
- Limits
- Up to 2 hours and 1 GB per file.
- Retention
- Audio is deleted after 7 days. Transcripts and summaries are deleted after 30 days.
- Deleting
- Delete a recording and everything derived from it yourself, at any time.
- Speech
- Azure AI Speech.
- Summaries
- Google Gemini.
- Interface
- English.
Questions
Questions people ask first.
- Is this a meeting bot?
- No. Nothing of ours joins your calls and there is no calendar connection. You upload a recording you already have, which is also why it works for an in-person interview.
- Which languages does it handle?
- Cantonese, English, and recordings that move between the two mid-sentence. The summary can be written in English or Traditional Chinese. The interface itself is English.
- How accurate is it?
- We do not publish an accuracy figure and we do not promise a correct transcript. Speech recognition gets things wrong, particularly names, numbers and people talking over each other. What Recolio does instead is show its working: where the speech engine reports low confidence in a segment, that segment carries a badge rather than a number, and every summary point links to the segment it came from so you can check it against the audio in one click.
- Who can see my recordings?
- Members of your workspace see only their own recordings. A billing owner sees usage totals, not the contents. Your audio goes to Azure AI Speech to be transcribed and the resulting transcript goes to Google Gemini to be summarized; those two are the processors involved.
- How do I delete a recording?
- There is a delete action on the result page that removes the audio and everything derived from it, and it cancels any work still running. If you do nothing, audio is deleted after 7 days and transcripts and summaries after 30 days.
- What happens after the trial?
- The trial covers 3 hours of audio and takes no card, so it simply runs out. Nothing is charged and nothing is deleted early. Subscribe a seat when you want to keep going.
Try it on a recording you already have.
3 hours of audio, no card. Audio is deleted after 7 days and transcripts and summaries after 30 days, and you can delete anything sooner.