Free Cantonese Speech-to-Text Options in 2026: 3 Tools Compared

Three common misconceptions to clear up before choosing a tool

Many people searching for "Cantonese speech to text" for the first time think, "I'll just install a shortcut and be done." But after trying it, they realize the bottleneck isn't the shortcut—it's these things:

• Misconception 1: Thinking "having a shortcut" means "having a transcript." A shortcut is essentially just a delivery mechanism; it sends the audio file to the recognition service. The real quality is determined by the engine and plan behind it.
• Misconception 2: Treating "free" as "unlimited." Free versions usually have limits on single-recording length, monthly minutes, or feature tiers. For example, cantonese.ai's official page states that speaker identification requires an upgraded plan. For what the free tier includes, refer to the current information on the official website.
• Misconception 3: Only comparing "whether it can transcribe" and ignoring what happens after transcription. Does it include names and timestamps? Can it segment? Can it generate summaries and action items directly? Can you ask a follow-up question like "Who mentioned the budget last time?"—these determine how much time you actually save.

First, clarify whether you want "a piece of audio turned into text" or "an entire meeting turned into usable data." The choice will be completely different.

Candidate list: positioning of the 3 free options

Tool	What you get for free	Platform
cantonese.ai	Cantonese speech to text, including iOS/macOS shortcut workflows; speaker identification requires an upgraded plan	Web, iOS Shortcuts, macOS Shortcuts, Android, Chrome extension
Tinrec	Free version offers basic quota to try recording, importing, summaries, and transcription	Mobile, desktop, web
Notta	Free 120 minutes per month, max 3 minutes per recording	Multi-platform and web

Below, we explain how each works and its limitations.

How to use the cantonese.ai shortcut: Turn WhatsApp voice messages into text

This is the most intuitive use case for this keyword, and the process is just three steps:

1. Get started: First, get the "Shortcut" provided by cantonese.ai on your iPhone and install it.
2. Action: Open WhatsApp, long-press the voice message you want to process, select "Forward," and in the iOS share sheet, find and tap "Speech to Text by cantonese.ai."
3. Result: Wait for the service to respond, and you'll get the transcript of that voice message.

If you have a meeting recording file instead of a voice message, cantonese.ai also offers web upload—just drop the audio file in to convert it to text. The official API documentation also mentions advanced options like timestamps and speaker diarization. Note that the synchronous endpoint is better suited for short audio files under 10 minutes; longer content should be processed asynchronously (submit + poll).

A few honest limitations upfront: the shortcut workflow is smoothest on iPhone; Android is supported but the steps differ from iOS; for free quota, maximum file length, and whether batch upload is supported, refer to what is displayed during actual use where the official page does not specify. Also, Cantonese accents, background noise, and overlapping speech will affect results, so important content should still be manually verified.

4 key points to understand before choosing a Cantonese transcription tool

1. Does it recognize "Cantonese" or "written Chinese"? Cantonese has many colloquial words and particles. Whether it can recognize them correctly depends on the tool's language training. Test it with a real recording of your own—that's more accurate than any promotional numbers.

2. What is the "shape" of the free quota? Is it total minutes per month, maximum length per recording, or file size limit? For example, Notta's free version offers 120 minutes per month but caps each recording at 3 minutes; this design is not friendly for "one long meeting"—you'll have to split it yourself.

3. What can you do after transcription? Timestamps, speaker segmentation, summaries, chapters, action items, AI follow-up questions, bilingual comparison—these determine whether you get "a piece of text" or "usable data."

4. How smooth is cross-device use and export? Recording on mobile and organizing on computer is common. If a tool only works on a single device, or the transcript can't be brought into workflows like Notion or Google Docs, you'll waste a lot of time copying and pasting.

Tinrec — If you need more than just a transcript

Tinrec is an AI meeting notes and collaboration tool for individuals and teams. It supports real-time recording, bot-free recording of online meetings on desktop, audio/video file import, and real-time translation. It can automatically generate summaries, chapters, action items, and mind maps, and even answer AI questions about meeting content.

During a meeting, it does two things at once. The desktop version captures computer system audio directly, handling online meetings on Zoom, Google Meet, Microsoft Teams, Webex, etc., without inviting an extra meeting bot to join as a listener. The transcript is generated in sync during the meeting, and after the meeting, summaries, chapters, and action items are already organized—who is responsible for what and when it's due, no need to re-listen to a two-hour meeting.

After recording, you can just ask questions. This is the biggest difference from pure transcription tools: you can ask questions about a single meeting or multiple meetings, such as "What was the conclusion of the last meeting?" or "What was the budget cap the client mentioned?" It doesn't return keyword search results but answers based on the content. For those who often need to look back at past meetings, this saves the entire time of searching.

The transcribed text should be portable. Transcripts, summaries, reports, and tables can be exported to Notion, Google Docs, OneNote, Dropbox, etc., for further processing; you can also use Agent post-processing to turn meeting content into reports or tables, directly as post-meeting deliverables.

Team scenarios have a separate team version. The team version is an independent team space, not a shared personal account: you can create multiple teams, invite members via join links, and distinguish owners, admins, and regular members; data belongs to the team, and resources can be handed over when members leave or lose their seats. Roles determine governance permissions, and seats determine whether one can record, upload, edit, and export—members who are active but have no seat can still view, play, and read data. There are also team hotwords, usage analytics, audit logs, and an audio recycle bin (currently retained for 30 days). Pricing: team monthly is USD 29.80 per paid seat per month, annual is USD 199 per paid seat per year (about USD 16.58 per month), each seat provides 2,000 minutes of team shared import quota per month, and eligible teams get a 7-day trial with 1 free seat and 300 minutes of shared quota; actual amounts and benefits are subject to the official purchase page.

Limitations should also be clear. The free version offers a basic quota to try; for actual minutes and plan details, refer to the official website; accuracy is affected by recording quality, noise, accents, overlapping speech, and technical terms, so it's not recommended to assume fixed numbers; the team version currently focuses on team-level sharing, not file-level or member-level granular permissions.

Who is it for? If you want to move meetings from "recorded" to "searchable, queryable, and portable," and often handle Chinese and multilingual meetings, Tinrec will be closer to your needs than a simple transcription service.

What other options are there? Who are Notta and Otter.ai for?

Notta: Multilingual transcription and file processing are its forte. Free: 120 minutes per month, max 3 minutes per recording; Pro annual billing is about USD 8.17 per month, with 1,800 minutes per month. Suitable for tasks focused on "file transcription + subtitles." However, long meetings on the free version require manual splitting, and Tinrec's bot-free meeting recording and post-meeting Agent output are areas it doesn't emphasize.

Otter.ai: A mature choice for English business meetings. Basic free: 300 minutes per month; Pro annual billing is about USD 8.49 per user per month, including 1,200 minutes of in-app recording and 10 file imports per month. If 90% of your meetings are in English, it's a good deal. But Chinese and Cantonese content are not its strong suit, and Tinrec's "record, generate action items, export to existing workflows" organization chain is something Otter focuses less on.

Pitfall guide: 4 most common traps when using free plans for Cantonese speech to text

Pitfall 1: Treating "free" as "unlimited." First check whether the limit is "minutes per month" or "minutes per recording." Plans with short per-recording limits will force you to split files for a one-hour meeting, and just splitting files can take half an hour.

Pitfall 2: Keeping only text, not audio and timestamps. Transcripts always have recognition errors, especially names, proper nouns, and numbers. If the tool doesn't keep audio and timestamps, verifying a sentence later will be very painful.

Pitfall 3: Ignoring technical terms. Company abbreviations, product codes, and job titles are often where errors are most consistent. Tools with hotword features can be fed these in advance to reduce back-and-forth corrections.

Pitfall 4: Recording in noisy environments with multiple people talking over each other. Even the best engine can't save that. For in-person meetings, keep the microphone close to the speaker and cultivate the habit of "one person speaks at a time"—this improves recognition quality more than switching tools.

One final practical reminder: recording and transcription involve personal data and local regulations. Notifying participants and obtaining consent before a meeting is basic courtesy and risk management.

Summary: Three needs, three choices

• Just occasionally converting WhatsApp voice messages to text → cantonese.ai shortcut. A few taps on your phone and it's done—the easiest entry point.
• Turning meetings into searchable, queryable data → Tinrec. Bot-free recording, summaries and action items, AI follow-up questions, multi-format export to team space—the whole chain in one go.
• Meetings almost entirely in English → Otter.ai's free 300 minutes can tide you over.
• Mainly file transcription and subtitles → Notta, but remember to avoid the 3-minute per-recording limit on the free version.

The recommended approach is simple: take your most realistic recording (the kind with background noise and accents) and run it through the free quota once. Compare the quality and how smooth the subsequent organization workflow is—you'll know at a glance, without paying upfront.

References

• Speech to Text - cantonese.ai
• cantonese.ai — Cantonese AI voice, speech and language tools
• API Documentation | cantonese.ai
• Cantonese Speech to Text Software | cantonese.ai
• [Tutorial] WhatsApp One-Click Voice Message to Text - Mobile Life