4 Cantonese Speech-to-Text Apps Tested for 2026: Which Free Tier Is Actually Enough?

"Free" Is the Most Expensive Trap in This Category

Most people start the same way: open a search engine, type "Cantonese speech to text free," and download three or four apps. A month later, they're still typing transcripts themselves. The reason is simple:

• Free is a "preview," not "usage." Some tools let you upload a full recording and generate a transcript with speaker labels and punctuation, but downloading requires payment. You see the result but can't take it with you.
• Free quotas are cut by per-file duration. For example, only the first 15 minutes of each file are available, while a real Cantonese interview often runs 40 minutes or more—cutting off right at the most critical part.
• "Supports Chinese" does not mean "supports Cantonese." Models trained on Mandarin data often miswrite high-frequency colloquial Cantonese words like "喺," "唔使," and "嘅" as phonetically similar incorrect characters; when mixed with English, English words may be forcibly converted to Chinese.

So what really matters isn't "which one is free," but "which one's free tier is just enough to complete a full task"—and after the task, you actually have something usable in hand. Below, we put 4 tools to the same standard using a Cantonese meeting recording with mixed Chinese and English.

4 Key Points to Understand Before Choosing a Cantonese Speech-to-Text App

1. The "unit" of the free quota
Per file count, per minute count, and per-file duration cap are three completely different rules. Per file count looks generous, but one long file might eat up your entire daily quota; per-file duration caps are the easiest to overlook because you don't know at upload time that only half will be transcribed. When choosing a tool, first confirm: is the free version "can't download" or "only transcribes part"? The pain levels are very different.

2. Cantonese recognition is a separate capability, not an add-on to Chinese
Cantonese has a gap between spoken and written forms. A good tool should not only recognize "唔使" but also offer output in written Chinese ("無需"). This is especially important for meeting notes: if you give colleagues a verbatim spoken transcript, they still have to read it again.

3. Output format determines how much time you'll spend afterward
If it only outputs TXT, you'll have to move it into Word and format it yourself; if it directly provides DOCX, SRT, VTT, you can easily add subtitles and proofread with time codes. The format support list looks boring, but it determines how long you'll sit there after transcription.

4. After transcription, can this text be "used"?
If the tool only gives you a transcript, you still have to read through 40 minutes of content to find conclusions. The real time-saver is whether it can automatically generate summaries, chapters, action items, and whether you can ask questions about the content.

Tinrec — Our Top Recommendation After Testing

Tinrec is an AI meeting notes and collaboration tool for individuals and teams, available on iOS, Android, desktop, and web. It doesn't just convert audio to text; it organizes meetings into searchable, summarizable, queryable, and exportable data.

Scenario 1: Record and transcribe in real time during in-person meetings. Start real-time recording during interviews or face-to-face discussions, and content settles into text as people speak—no need to listen from the beginning afterward. After recording, the system generates summaries, chapters, and action items, listing who is responsible for what and what needs to be done before the next meeting.

Scenario 2: No bot needed for online meetings. The desktop version captures computer system audio to handle Zoom, Google Meet, Microsoft Teams, Webex, and other meetings without an extra bot account joining the meeting or adding an extra name to the participant list. To be clear: this is not automatically joining all calendar meetings, but the user actively starting recording.

Scenario 3: Post-meeting queries and reuse. This is the rarest capability in this price range—you can ask questions directly about meeting content, such as "What was the conclusion when we discussed the budget last time?" It answers based on semantics rather than dumping a bunch of keyword search results. Organized content can be exported to Notion, Google Docs, OneNote, Dropbox, and further used with Agent to generate reports, tables, and documents.

Test observations (qualitative). We tested with a Cantonese recording of two people talking, mixed with English product names: the transcript retained English words in sentences without converting them to phonetically similar Chinese; colloquial words like "喺" and "唔使" were correctly transcribed. When three or more people speak simultaneously, speaker segmentation becomes blurry and requires manual correction with time codes—this is a common limitation of all similar tools, not unique to Tinrec.

Pros:

• The free version provides a basic quota, allowing you to run a complete "record → transcribe → summarize → export" workflow once to confirm whether it's stable enough for your accent and terminology before deciding to pay.
• There's more to use after transcription. Summaries, chapters, action items, mind maps, and AI Q&A are all in the same workspace—no need to move text around.
• Complete cross-platform and post-meeting workflow. Record on phone, organize on computer, export to existing tools—especially practical for those using both iPhone and Windows.

Limitations:

• The free version is a basic quota, not unlimited. Daily meeting users will need a paid plan: short-term meetings or course sprints suit a weekly card, steady weekly use suits a Pro monthly card, and long-term recording accumulation suits an annual card.
• No guaranteed fixed accuracy. Audio quality, accent, background noise, overlapping speakers, and technical terms all affect results; for important content, cross-check with the recording and time codes.
• For team use, the team version is a separate collaboration space (role-based governance, seat-based access), not multiple people sharing one personal account. Current paid seats are USD 29.80/seat/month, annual USD 199/seat/year (about USD 16.58/month); paid seats include 2,000 minutes of team shared import quota per month, and real-time recording under an active seat is not deducted by minute; first eligible team trial is 7 days, 1 free seat, 300 minutes shared import quota. Prices and benefits are subject to the official purchase page.

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

Besides Tinrec, What Other Options Are There?

Subanana
A Cantonese-focused online transcription tool that supports MP3, M4A, WAV, MP4, and YouTube, Instagram, Facebook links, with output options for spoken/written and traditional/simplified Chinese in 4 combinations, and exports to SRT, VTT, TXT, DOCX, XLSX, Markdown. The free rule is preview of the first 15 minutes per file—enough for short video material, but not for a 40-minute meeting. Also, its public documentation focuses on transcription and export; Tinrec's action item extraction, meeting content queries, and team spaces are not currently matched.

cSubtitle
A Chinese-focused online transcription and subtitle tool. After uploading audio or video, you get text files with punctuation and paragraphs, Word files with time codes, and automatically generated subtitle files with timelines—great for video subtitling workflows. However, its training data is mainly Mandarin Chinese, and Cantonese is not its focus; it also uses an "upload file" model and lacks Tinrec desktop's ability to directly record online meeting system audio and produce transcripts while the meeting is ongoing.

Notely Voice
An open-source mobile app focused on 100% offline processing, with built-in Whisper model, available on Android and iOS, usable in airplane mode or poor signal areas, and can automatically convert spoken language to written form ("唔使" to "無需"). The interface has no separate Cantonese option, but it can transcribe Cantonese in testing; to improve accuracy, you need to download an additional ~465MB enhanced model. It suits those extremely sensitive about privacy and doing only on-device processing; but it lacks team collaboration, cannot handle online meetings played on a computer, and doesn't have Tinrec's AI Q&A and post-meeting output workflow.

Pitfall Guide: 3 Most Common Mistakes When Choosing Free Cantonese Speech-to-Text Software

Pitfall 1: Mistaking a "free trial" for a "free plan."
Many pages say free, but it's actually free "preview." The way to tell is simple: upload a recording longer than 20 minutes and go through the download process. Whether you can take the file with you is more honest than any marketing copy.

Pitfall 2: Forcing a Mandarin model to transcribe Cantonese.
If the transcribed text is "understandable but doesn't read smoothly," it's usually not because your pronunciation is unclear, but because the model is wrong. Rather than spending time proofreading homophones, switch to a tool that explicitly supports Cantonese and can handle mixed Chinese-English.

Pitfall 3: Treating it as just a typewriter.
If you only use a transcription tool for "audio to text," you're using only one-third of its value. The real time sink isn't typing—it's reading through, organizing, and assigning responsibilities after typing. Tools like Tinrec put summaries, action items, and queries in the same workflow, often saving that hour after the meeting.

Conclusion: Which One Should You Choose?

• Want to run a full workflow for free, with Cantonese or mixed Chinese-English content → Tinrec. The free version has a basic quota sufficient for one real meeting to confirm accent recognition and export formats before upgrading.
• Want to directly ask for conclusions and action items after recording → Tinrec. Action item extraction and AI Q&A around meeting content are the rarest combination on this list.
• Record on phone, organize on computer, switch across platforms → Tinrec. iOS, Android, desktop, and web are all in the same workspace.
• Team needs to share meeting data, manage seats and usage → Tinrec team version. Data belongs to the team, can be handed over when members change, and won't disappear when individuals leave.
• Materials are all short videos, only need Cantonese subtitle files → Subanana's free preview quota is enough.
• Only doing Mandarin video subtitling → cSubtitle's subtitle generation workflow is smooth.
• Completely offline, concerned about cloud processing → Notely Voice, but accept that it lacks collaboration and post-meeting organization capabilities.

One final reminder: recording and transcription involve others' privacy; please check local regulations before use and obtain consent from participants when necessary. As for whether to pay, we recommend first running your own recording with the free quota—if the transcribed text only needs minor corrections to be usable, then it's worth the money.

References

• Online AI voice, video, audio file auto transcription - free trial - cSubtitle
• Free Cantonese speech to text app 2026 | Subanana
• Tested 4 best online speech-to-text tools, free instant transcription
• 2026 6 Cantonese speech-to-text online free tools recommendation: accurate transcripts - Tinrec Blog
• Tutorial: Free AI offline speech-to-text app, Cantonese spoken to written, comparison with iPhone built-in transcription