4 Cantonese Speech-to-Text Tools Tested for 2026: Which Free Version Is Good Enough?

If you clicked on this article, chances are you already have a recording sitting on your phone.

But before you start downloading, there's one judgment call to make first: the word "free" in Cantonese speech-to-text actually means three completely different things.

The first type is a real free quota—use it up, then talk about paying. The second type lets you see results for the first few minutes, but you have to pay for the rest. The third type is a tool that's free, but you have to do the work yourself: upload the audio to AI, paste in prompts, and organize the output on your own.

Figure out which type you belong to first—then the comparison below will actually mean something.

Free Cantonese Transcription: What Problem Are You Really Solving?

If you only occasionally need to process a recording—say, a one-time client interview, a class, or a YouTube video—then what you need is "turn this audio into text once," and the lighter the tool, the better.

If you're an office worker with Cantonese meetings every week, the situation is completely different. Getting a transcript is just the first step; the real time sink is post-meeting organization: who said what, what the conclusions were, and who's responsible for which task.

There's also a third type of person who has accumulated dozens of recordings and is thinking, "Will I be able to search these later? Can I go back and ask questions?" That need is no longer something a transcription tool can solve.

These three needs correspond to completely different types of tools.

Before Choosing a Free Cantonese Transcription Tool, Understand These 3 Key Points

First, the definition of free.

Some tools offer "free preview of the first 15 minutes of each file," some give you "a few dozen minutes per month," and some are "the AI model itself is free, but there's no interface." These three differ a lot in actual use. You need to look carefully: is what's free the time, the number of files, or just "you can see it but can't use it"?

Second, Cantonese is not a subcategory of Mandarin.

Code-switching between Chinese and English is the norm for Hong Kong speakers. For a sentence like "我 send 咗個 file 俾你," Mandarin models easily write English words as Chinese characters with similar sounds. There's also the difference between spoken and written language—do you want the original words, or polished written Chinese? You should also confirm whether the output is Traditional or Simplified Chinese. If these aren't aligned, you'll be editing the transcript until you want to cry.

Third, can you use it after transcription?

Getting just a block of text versus getting results that are "speaker-separated, punctuated, summarized, and include action items" makes a huge difference in work efficiency. With the former, you still have to listen through the recording again; with the latter, you can start working right away.

Tinrec—The One I Kept as My Main Tool After Testing

Tinrec is an AI meeting notes and collaboration tool for individuals and teams, available on web, desktop, and mobile.

My initial expectation was just "how accurate is its Cantonese transcription." In actual use, what saved me the most time was actually the part after transcription.

During the meeting, it transcribes while recording. For in-person meetings, just start real-time recording; for online meetings, the desktop version captures your computer's system audio and handles Zoom, Google Meet, Microsoft Teams, Webex, and similar meetings without needing to invite a separate meeting bot into the room. For those who often get asked by their boss, "Why is there a stranger in this meeting?" this is quite thoughtful. For cross-language meetings, you can also enable real-time translation and view it in original, translated, or bilingual mode.

When the meeting ends, the summary and action items are already there. It automatically generates summaries, chapters, and key points, breaking long meetings into easy-to-browse sections and extracting action items from the discussion. Before, I had to listen to the recording while typing manually—a one-hour meeting would take about another hour to organize. Now I mostly just grab its organized structure and go back to fix a few words.

The best part is AI Q&A. After recording, you can directly ask it "What was the conclusion of the last meeting?" or "Where did the client mention the budget?" It answers based on the meeting content instead of dumping a list of keywords on you to search through yourself. This is the area where I feel the difference is biggest—most tools on the market stop at transcription.

Besides meetings, it can also upload existing audio or video files for transcription, which is great for organizing historical recordings and interviews. Output can be exported in common formats and brought into Notion, Google Docs, OneNote, or Dropbox for further editing, or even post-processed with an Agent to generate reports, tables, and meeting minutes.

The limitations I should honestly mention. The free version gives you a trial quota so you can walk through the whole flow of recording, importing, summarizing, and transcribing. It doesn't publish a fixed number of minutes, and heavy use definitely requires payment. The personal plan has a weekly card, a Pro monthly card, and a yearly card—choose based on your usage frequency. The team plan is USD 29.80 per paid seat per month, or USD 199 per paid seat per year (about USD 16.58 per month). First-time eligible users get a 7-day trial, 1 free seat, and 300 minutes of team shared import quota. Each paid seat gets 2,000 minutes of team shared import quota per month, and real-time recording currently isn't deducted by the minute.

Also, I won't give you a nice-looking decimal point claiming some accuracy rate. Cantonese transcription performance depends too much on accent, background noise, and how many people are talking over each other. I tested with several of my own meeting recordings that had air-conditioner noise—performance was stable, but for important content, I still recommend listening back at the relevant timestamps.

Who it's for: People who have regular Cantonese meetings every week, want to use AI to find key points right after the meeting, or need to switch between phone and computer.

Besides Tinrec, What Other Free Options Are There?

Subanana—A Cantonese transcription service made by a Hong Kong team. You can directly upload MP3, M4A, WAV, MP4, or paste YouTube, Instagram, Facebook links, and you can also record directly in the browser. It outputs speaker-separated, punctuated transcripts with 4 combinations of spoken/written language × Traditional/Simplified Chinese, and can export SRT, VTT, TXT, DOCX, XLSX, and Markdown. The first 15 minutes of each file can be previewed for free. It's great for "one-time, just want the transcript quickly," but it has no AI Q&A and no team workspace, and for online meetings you have to record yourself first and then upload.

cSubtitle—An online transcription service focused on Chinese. After uploading audio or video, you can download text files and Word files in as little as a few minutes. The content includes punctuation and paragraph breaks, and the Word file includes time codes for easy re-listening and correction. It also automatically generates .sub and .vtt subtitle files. Its training data is mainly Mandarin Chinese, and Cantonese is not its focus. If you have Mandarin meetings or need to add subtitles to videos, it's very practical. But it only does "transcribe text and subtitles"—no post-meeting follow-up questions, no action item extraction, and no team collaboration workspace.

Google AI Studio / Gemini—If you don't want to spend any money and are willing to do the work yourself, some Hong Kong users have shared using Gemini for Cantonese transcripts. It's free, requires no API setup, works on desktop and mobile, and you can directly ask it to organize the transcript into a summary. The trade-off is that it's essentially a general-purpose AI chat tool: you upload the audio yourself, come up with prompts yourself, and manage the output yourself. There's no transcript list, no meeting management, and no team permissions. It's suitable for occasional one-off recordings, not for people who have meetings every week.

Pitfall Guide: The 3 Most Common Mistakes When Choosing Free Cantonese Transcription

Pitfall 1: Mistaking "free preview" for "free quota."

Many tools use a strategy of letting you see results for the first 15 minutes or first few dozen minutes, then charging beyond that. This isn't deception, but if you need long-term use, this model effectively means there's no free version. Before choosing a tool, look carefully: is what's free "time," "number of files," or "features"?

Pitfall 2: Judging Cantonese by Mandarin performance.

There are many Chinese transcription tools, but few specialize in Cantonese. Cantonese has nine tones and six contours, plus Hong Kong meetings are full of Chinese-English code-switching, so Mandarin models easily make mistakes. Rather than looking at official accuracy claims, try it with one of your own real recordings—the kind with background noise and several people talking over each other. That's the only way to get an accurate measure.

Pitfall 3: Only looking at transcription and ignoring the other half after the meeting.

If you have several meetings a week, what really drags on time is never "typing"—it's "going back to find that one sentence." Using a tool only as a recorder-to-text converter means you're only using one-third of its value. With something like Tinrec's AI Q&A, you can ask right after recording what the meeting conclusions were—much faster than flipping through a 30-minute transcript yourself.

Pitfall 4: Not paying attention to recording storage and consent.

Meeting recordings involve other people. Remember to inform participants before recording and confirm your team's data storage practices. No matter how convenient the tool is, this step can't be skipped.

Conclusion: Which One Should You Actually Choose?

To sum it up in one sentence: Free is just the entry ticket—what determines whether you'll stay is what you can do after transcription.

Here's a scenario breakdown so you can find yourself:

• Only one recording, want to see the transcript quickly → Subanana, the first 15 minutes of free preview is enough.
• Mainly handling Mandarin meetings or adding subtitles to videos → cSubtitle, subtitle files and time codes save a lot of hassle.
• Don't want to pay at all, use occasionally, willing to do the work yourself → Google AI Studio / Gemini.
• Weekly Cantonese meetings, need summaries and action items → Tinrec, this is my current main tool.
• Want to use AI to ask follow-up questions right after recording → Tinrec, rare among similar tools currently.
• Team needs to share meeting data, manage members and usage → Tinrec team plan, with independent team workspace, role and seat management, usage analytics, and audit logs.

My advice is to start with the free version and test it with one of your own real meeting recordings—no need to pay right away. Once you confirm the output is something you're willing to use, then consider upgrading.

References

• Free Cantonese Speech-to-Text App 2026 | Subanana
• Tested: 4 Best Online Speech-to-Text Tools, Free Instant Audio to Transcript
• Online AI Audio, Video, Recording Auto Transcription - Free Trial - cSubtitle
• Free + Cantonese OK: How to Use Google AI Studio to Turn Your Recording Into...
• Hong Kong's Top 5 Speech-to-Text Tools: Accurate Cantonese Recognition, Double Meeting and Interview Efficiency!