4 Free Cantonese Speech-to-Text Apps Tested: Which One Has the Most Usable Free Tier in 2026?

The word "free" is the easiest way to waste your time

Many people looking for a free Cantonese speech-to-text app follow this process: open the App Store, search for "speech to text," see "free," and download. After recording a one-hour meeting, they discover the free version only gives them 3 minutes.

Or another scenario: the tool is indeed free, but the transcribed text is full of errors. Because it forces a Mandarin model onto Cantonese, "m goi" (thank you) becomes "bat goi" (should not), and "hoi wui" (meeting) is heard as "hoi fui" (open ash). Proofreading takes longer than typing it out yourself.

There's an even more common misunderstanding: thinking that "getting a transcript" is the end. You end up with a 30,000-word text file with no paragraphs, no speaker labels, and no timeline. To write meeting minutes, you still have to read through it from the beginning.

This article isn't another feature list. It's here to help you avoid these pitfalls: how to calculate free quotas so you don't lose out, whether Cantonese is truly being recognized, and the step after recording that actually saves time.

Understand before choosing: 4 key points for free Cantonese speech-to-text apps

Key 1: Is the free quota in "minutes" or "files"?

They all say free, but the calculations vary widely. Some offer free preview of the first 15 minutes per file, some give a fixed number of minutes per month, and some limit each file to 3 minutes.

The most practical way to judge is to plug in "the length of a meeting." Someone who has two one-hour meetings per week versus someone who transcribes one interview per month will need completely different plans.

Key 2: Does it use a Cantonese model or a Mandarin model?

This is the easiest thing to hide. Many tools do have "Chinese" in the language menu, but they actually run a Mandarin model. When they encounter colloquial Cantonese, mixed Chinese-English, or Hong Kong-specific terminology, they start guessing wildly.

The test is simple: take a recording of your own everyday speech, preferably including a few mixed Chinese-English sentences. Can it segment sentences correctly? Can it retain colloquial words like "hai" (at) and "m sai" (no need)? You'll know right away.

Key 3: Can the output format be used directly?

The value of a transcript isn't in "having text" but in "not having to organize it again." Things to check include: automatic punctuation, speaker separation, a timeline for cross-referencing, and export to Word, SRT, TXT, or Markdown.

If you just need to write an article, plain text export is enough. But for subtitles or meeting minutes, timeline and speaker labels become essential.

Key 4: After recording, how much does the tool do for you?

This is where the gap has widened the most in recent years. Some tools only give you a transcript, while others go a layer further: automatic summaries, extracting action items, and letting you ask questions directly about the meeting content.

For the same one-hour meeting, one version requires you to read it all yourself, while the other lets you directly ask "What was the conclusion of this meeting?" or "Who is responsible for the budget?" The time cost differs by several times.

Tinrec: After actually trying it, this is the one I recommend most often

Tinrec is an AI meeting notes and collaboration tool for individuals and teams, available on web, desktop, and mobile. It handles Chinese, Traditional Chinese, and multilingual meetings. It doesn't just convert audio to text; it organizes meetings into data that can be searched, queried, and further processed.

No bot needed for online meetings. This resonates especially with users in Hong Kong and Taiwan. Most AI meeting tools work by sending a bot account into the meeting, adding an extra participant on screen, which some clients find odd. Tinrec's desktop version directly captures the computer's system audio, so when handling meetings on Zoom, Google Meet, Microsoft Teams, or Webex, no bot needs to join. Note that this is not "automatically joining all calendar meetings"; you open Tinrec yourself to start recording.

Transcribe while recording, have something to review when the meeting ends. For offline interviews or in-person meetings, use the real-time recording feature to transcribe as you record. Afterward, AI generates summaries, chapters, and key points, and organizes action items from the discussion into to-dos. For cross-language meetings, you can enable real-time translation and view the original, translation, or bilingual side-by-side.

Ask it directly after recording. This is what I find most useful and most easily overlooked. Instead of using Ctrl+F to find keywords in a one-hour transcript, you can ask Tinrec directly: "What was the budget cap the client mentioned?" or "What was decided in the last meeting?" It answers based on the meeting content, not a bunch of keyword search results.

Bring results back to your existing workflow. Organized content can be exported in various file formats and sent to Notion, Google Docs, OneNote, or Dropbox for further editing. When you need formal documents, you can use Agent post-processing to generate reports, tables, or meeting minutes.

For team use, Tinrec's team version is a separate collaboration space, not multiple people sharing one personal account. It centralizes meeting data within the team, offering member invitations, roles (owner/admin/member), seat allocation, usage analytics, audit logs, audio recycle bin, and team hotwords. It's important to distinguish "roles" from "seats": roles determine who can manage the team, while seats determine who can record, upload, edit, and export. Members without a seat can still view, play, and read team data.

Current pricing for the team version: eligible teams can try free for 7 days with 1 free seat and 300 minutes of shared team import quota. Monthly billing is USD 29.80 per paid seat per month; annual billing is USD 199 per paid seat per year (about USD 16.58 per month), with each paid seat providing 2,000 minutes of shared team import quota per month. Under an active plan and active seats, real-time recording currently does not deduct minutes; file and web imports use the shared team quota. Actual prices and benefits are subject to the official purchase page.

I think it's worth recommending for three reasons: First, the "no bot" aspect is crucial in real client meetings—it doesn't make the other party feel the meeting is being additionally recorded. Second, the post-meeting part is very complete: summary, to-dos, Q&A, and export form a whole workflow, not scattered features. Third, team data has ownership—when members leave, data stays with the team and doesn't disappear with personal accounts.

Limitations to be honest about: The personal free version offers a basic quota, suitable for trying out a meeting to get a feel for it; long-term heavy use still requires evaluating a paid plan. The team version currently focuses on team-level sharing, where all active members can read team data—it's not fine-grained file-level permission control. Transcription accuracy is affected by recording quality, background noise, accents, overlapping speech, and technical terminology. It's not recommended to treat it as "completely proofreading-free"; important content should still be verified by listening back with the timeline.

Who it's for: People who frequently handle Chinese or mixed Chinese-English meetings, want summaries and to-dos right after recording, need to switch between phone and computer, or want to turn meeting data into team assets.

Besides Tinrec, here are other free options

Subanana: A Cantonese speech-to-text service developed by a Hong Kong team, featuring a Cantonese model. You can upload audio files, videos, or paste YouTube links for transcription. It automatically separates speakers and adds punctuation, offers colloquial and written Chinese options, and handles mixed Chinese-English well. Each file's first 15 minutes can be previewed for free, suitable for verifying results first. However, it's mainly a web tool and lacks Tinrec's desktop bot-free recording for online meetings, as well as team roles and seat governance.

cSubtitle: An online audio/video-to-text service with Cantonese (HK) in the language menu. No account registration required for uploads. Supports MP4, MOV, MP3, WAV, M4A, etc., with a single file limit of about 4GB and up to 5 hours. Output includes punctuation, paragraphing, and a Word file with time codes plus subtitle files. The free version is a quick trial for short durations. Its strength is converting existing files to subtitles; real-time recording, AI Q&A, and post-meeting to-dos are areas Tinrec covers.

Notta: A multilingual transcription tool with good Chinese support and team plans. The free version offers 120 minutes per month with a 3-minute limit per session. Pro annual billing is about USD 8.17 per month for 1,800 minutes per month. The 3-minute per-session limit is quite tight for meeting notes. It overlaps significantly with Tinrec, but if you want desktop system audio capture for bot-free recording, AI Q&A around meetings, and team management like members/seats/audit, Tinrec's approach is closer to meeting collaboration.

(Prices and benefits of competitors above are subject to change; please refer to official pages before use.)

Pitfall guide: 5 most common mistakes when choosing free Cantonese transcription

Pitfall 1: Only seeing "free" without checking the quota unit. "N minutes per month," "3 minutes per session," and "first 15 minutes per file" are three completely different things. Testing with a recording of your actual meeting length is more useful than reading a hundred lines of specs.

Pitfall 2: Forcing a Mandarin model onto Cantonese. "Chinese" in the language menu doesn't mean it understands Cantonese. When testing, deliberately say a few colloquial sentences, like "m goi bong ngo tai ha" (please take a look for me) or "ni go fong on ma ma dei" (this plan is so-so), and see if the segmentation and word choices make sense.

Pitfall 3: Ignoring the difference between colloquial and written Chinese. If you want a faithful transcript, keeping colloquial speech is correct. If you want to publish directly or write a report, being able to switch to written Chinese saves a lot of editing time. Think about this before choosing a tool.

Pitfall 4: Only taking the transcript and wasting the AI features. Using a transcription tool only as "audio to text" means using only part of its capability. After recording, directly asking "What's the conclusion?" or "What are the to-dos?" is often much faster than manually scanning the transcript.

Pitfall 5 (most common in teams): Multiple people sharing one account. Meeting data gets mixed into personal accounts, and when someone leaves, files can't be found. If meetings are shared by the team, choose a solution with team spaces, roles, and seats, so data belongs to the team and can be handed over when members change.

Conclusion: Which one should you choose?

Free tools aren't unusable; you just need to pick the right "free" format. Brief conclusion:

• For meetings over an hour, wanting summaries and to-dos right after recording → Tinrec (real-time recording + AI summary + to-do extraction)
• Want to ask questions directly about the content after recording, without manually scanning the transcript → Tinrec (AI Q&A is its core differentiator)
• Don't want a bot joining online meetings → Tinrec (desktop captures system audio)
• Team sharing of meeting data, needing member and seat management → Tinrec team version (spaces, roles, usage, audit)
• Only need to convert existing videos to subtitle files → A file transcription service like cSubtitle is enough
• Only need to verify recognition of a certain Cantonese recording → Try Subanana's free preview first

My advice is to start with the free version and test with a recording of a real meeting. The key isn't "how many features" but "how long you need to edit the transcribed text" and "how much time post-meeting organization takes." If you save time on these two things just once, the tool's value pays for itself.

Finally, a reminder: recording and transcription involve others' privacy. Under local regulations, you should obtain consent from participants before proceeding.

References

• Free Cantonese Speech-to-Text App 2026 | Subanana
• Top 5 Speech-to-Text Tools in Hong Kong: Accurate Cantonese Recognition, Double Meeting and Interview Efficiency!
• Tutorial: Free AI Offline Recording-to-Text App, Cantonese Colloquial to Written Chinese, Compared with iPhone Built-in Transcription
• Tested: 4 Best Online Speech-to-Text Tools, Free Instant Audio to Transcript
• Online AI Speech, Video, Audio File Auto-Transcription - Free Trial - cSubtitle