2026 Comparison of 5 Cantonese Speech-to-Text Tools: Can You Use the Transcripts Directly?

Why Do You Need a "Cantonese Speech-to-Text File"?

Most people's first misconception about this is: throw a Cantonese recording into a tool, and you'll get a meeting minutes document ready to send out.

I used to think that way too.

Later I realized that what comes out is usually just raw material. It might be a long string of text without punctuation, two people's speech blended together, "m goi" written as something else, and most commonly, sections with mixed Chinese and English completely garbled.

So the real meaning of "Cantonese speech-to-text file" is not "turning sound into words," but "turning a recording that you can understand but is hard to use into a document that can be edited, searched, and reused."

It usually appears in three forms:

• Transcript (TXT, DOCX): Preserves every sentence completely, suitable for interviews, meetings, and research records.
• Subtitle file (SRT, VTT): With time codes, suitable for adding subtitles to videos.
• Structured file (CSV, XLSX): Suitable for organizing multiple recordings into tables or databases.

The difference is never about "whether it can transcribe," but about "how much time you still need to spend redoing it after transcription."

In this article, I put several common solutions together and clarify who each is suitable for.

Tinrec — Turning Recordings into Data You Can Continue Working With

If what I'm dealing with is not just a single Cantonese recording, but a meeting, an interview, or a whole set of research data that needs to be preserved, I would open Tinrec first.

Its positioning is not a "transcription tool," but an AI meeting notes and collaboration tool. This difference gradually becomes apparent in actual use.

The first difference is whether the input sources are diverse enough. Real-time recording allows transcription while recording, suitable for offline meetings and on-site interviews; the desktop version directly captures computer system audio, so when handling online meetings like Zoom, Google Meet, Microsoft Teams, and Webex, you don't need to invite a meeting bot to join as a listener; existing audio and video files can be uploaded directly; the mobile app handles on-the-go and temporary notes.

The second difference is that there's more after transcription. The transcript is just the starting point; it then generates summaries, chapters, key points, and action items, and you can continue to use AI Q&A to ask follow-up questions about a single meeting or across meetings, and even process the content into reports, tables, and documents. This, I believe, is the most fundamental distinction between it and "pure transcription" tools.

The third difference is the output. Results can be exported in common formats and brought into Notion, Google Docs, OneNote, Dropbox for continued use, rather than being stuck in a certain tool.

The fourth difference is in team scenarios. The team version is an independent collaboration space, not multiple people sharing a personal account. Meeting data belongs to the team, and you can create multiple team spaces, invite members via join links, set owners/admins/regular members, and assign or revoke seats. Here's a detail that is often misunderstood: roles determine who can manage the team, seats determine who can record, upload, edit, and export. Members who are active but have no seat can still view, play, and read team data, but cannot perform actions like uploading, editing, and exporting. There are also usage analytics, audit logs, audio recycle bin, and team hotwords.

Regarding pricing and quotas, the personal version has a free version, weekly card, Pro monthly card, and Pro annual card—just choose according to your usage pace; the team version offers a 7-day trial for first-time eligible users, 1 free seat, and 300 minutes of team shared import quota, with subsequent monthly payment at USD 29.80 per paid seat per month, or annual payment at USD 199 per paid seat per year, with each paid seat providing 2,000 minutes of team shared import quota per month; under an active team plan and active seats, real-time recording is currently not deducted by minutes. Actual amounts and benefits are subject to the official purchase page.

Who it's for: People who need Chinese and multilingual meeting notes, need to organize them into minutes and action items after the meeting, and want data to stay in the team space rather than scattered across personal devices.

An honest reminder: Cantonese is highly colloquial, with frequent mixing of Chinese and English, and when there's noise and multiple people talking over each other, any tool's recognition quality will be affected. My suggested approach for Tinrec is to first run a segment of your own real recording to see what the output looks like, and then decide whether to hand over the entire batch of data to it.

Subanana — A Transcription Tool Designed for Cantonese

If what you need to handle is pure Cantonese content, Subanana is a very intuitive choice. Upload a recording file or video, or paste a YouTube link, and it will output a transcript with speaker separation and punctuation added. Handling mixed Chinese and English is its main strength, and you can choose between colloquial or written style, and specify Traditional or Simplified Chinese. It can export SRT, VTT, TXT, DOCX, XLSX, and Markdown. It also supports direct recording in the browser, custom vocabulary lists, AI summaries, and asking questions about the recording.

Free accounts can upload and preview the transcription results for the first 15 minutes of each file, no credit card required, but exporting files is a paid feature. The official documentation also explicitly states that accuracy will decrease in noisy environments or when multiple people talk over each other, and recommends using the free preview first to confirm your recording quality.

Who it's for: Content creators and researchers who primarily work with Cantonese and need multi-format output (especially subtitles and tables).

UniScribe — General Transcription with the Broadest Format Support

UniScribe takes a general-purpose approach, with a long list of supported file formats: mp3, mp4, mpeg, mpga, m4a, wav, webm, mov, aac, ogg, opus are all supported. It offers 6 output formats, including srt, vtt, docx, txt, pdf, and csv, and can also automatically generate summaries and mind maps, suitable for quickly grasping the key points of long content. It uses an optimized Whisper large model, and the official statement says 1 hour of audio can be transcribed in about 1 minute, but also notes that final quality depends on the cleanliness of the audio itself.

Who it's for: People with files in various formats who need to export multiple formats at once.

cSubtitle — Batch Processing and Subtitle Files

If your volume is large, cSubtitle's batch processing will save you effort. In addition to standard text files and Word transcripts, it automatically adds punctuation, segments the text, and generates subtitle files with time codes. Its approach is very practical: meeting notes, interviews, courses, speeches, and documentation of audiovisual materials all require transcripts, and manually pausing and playing repeatedly is very time-consuming.

Who it's for: People who have a batch of recordings to process at once and whose final product is subtitles or Word documents.

CantoSubAI — Detail-Oriented with Cantonese Branches and Dictionary Settings

CantoSubAI's feature is that it handles differences within Cantonese in more detail, supporting Hong Kong Cantonese, Guangzhou dialect, Dongguan dialect, Shunde dialect, and other branches, and provides personalized dictionary settings that can convert colloquial Cantonese into written Chinese and output in Traditional Chinese. It supports MP4, MOV, MP3, WAV, and other formats, and can export txt, docx, pdf.

Who it's for: People whose recordings have obvious regional accent differences, or who need to rewrite a large amount of colloquial speech into written documents.

SoundType AI — Mobile Recording and Transcription on the Go

SoundType AI provides transcripts, AI summaries, and speaker identification in the form of a mobile app, supporting over 90 languages, suitable for transcribing anytime, anywhere after recording. Its Cantonese performance has some actual feedback in user reviews: some think the conversion success rate is about 85–90%, "acceptable but not enough," while others report that the free trial time is short, long recordings contain irrelevant information, and multi-speaker recognition and segmentation could be better. These are subjective user experiences, and actual performance still depends on your recording conditions.

Who it's for: People who mainly record with their phones, need instant transcripts and summaries, and can accept the limitations of an app.

With So Many Options, How Should You Choose?

My own decision order is like this:

• You want data that can be used directly after the meeting, not just a transcript → Tinrec, because summaries, chapters, action items, and AI Q&A are completed in the same workflow.
• You're recording an online meeting and don't want a bot to join the meeting room → Tinrec desktop version captures system audio, which is especially useful for external client meetings.
• Meeting data needs to stay with the team, and members may change → Tinrec team version's spaces, roles, seats, and data handover mechanisms are designed for this scenario.
• It's purely about adding subtitles to Cantonese videos → Tools like Subanana or cSubtitle that focus on subtitle output are more direct.
• File formats are mixed, and you need multiple export formats → UniScribe.
• Recordings have obvious regional accents → CantoSubAI's Cantonese branches and dictionary settings are worth trying.
• You just want to jot down a note on your phone → An app like SoundType AI is enough.

If you're still undecided, my advice is: first think about "who will this text file ultimately be given to, and what will it be used for," and then go back to choosing a tool. Doing it in reverse usually leads to extra detours.

4 Things to Know Before Using Cantonese Speech-to-Text

1. First, distinguish whether you need a transcript or a subtitle file. These two outputs have different purposes and different formats. If you don't think it through at the beginning, you'll have to run it again later.

2. Colloquial and written language are two different products. If Cantonese colloquial speech is written out directly, many people find it hard to read; converting it to written language means an extra layer of rewriting. Decide which one you need in advance.

3. For free quotas, check clearly whether you can "view" or "take." Some tools allow free preview of the first 15 minutes of results, but exporting requires payment; some free plans have very short single-session durations. Rather than comparing free minutes, it's better to confirm whether the "export" step is included.

4. Instead of chasing accuracy numbers, build your own terminology list. Names of people, company names, product codes, and industry terms are always where the error rate is highest. For example, Tinrec team version has team hotwords, and Subanana and CantoSubAI also provide custom vocabulary or dictionary settings. Building them first before running makes a big difference.

One more thing that's outside the scope of tools but important: recording and transcription involve other people, so please handle it in accordance with local regulations and obtain consent from participants when necessary.

Conclusion

"Cantonese speech-to-text file" is not a single file, but a process: recording goes in, text comes out, and then it's organized into something actionable.

If your need is only "turning sound into words," many tools on the market can do that—just pick one that feels comfortable.

But if your need is actually "after this meeting I still need to write minutes, follow up on action items, and look back to check who said a certain sentence," then it's worth adding Tinrec to your test list—it handles exactly the part after transcription.

You don't need to use all the features at once. Start with a real recording, run it once, see if the output can save you time reorganizing, and then decide the next step. Test gradually, and you'll slowly master this method.

References

• Free Cantonese Speech-to-Text App 2026 | Subanana
• Free Transcription of Chinese (Cantonese) Audio in Seconds
• Online AI Speech, Video, and Audio File Auto Transcription - Free Trial - cSubtitle
• AI Cantonese Speech-to-Text Professional Cantonese Recognition
• SoundType AI Ultra-Accurate Transcripts, Speech-to-Text, AI Dictation - Google Play App