The Language Gap

If you speak English, you have plenty of transcription options. Every tool supports English. Apple Intelligence handles it well. No complaints.

But if you speak Finnish, or Thai, or Swahili, or Catalan? Your options shrink dramatically. Most transcription tools either don't support your language at all, or support it so poorly that the output is unusable.

This is the language gap in voice-to-text technology. The majority of the world's languages are underserved or ignored entirely.

OpenAI's Whisper model changes this. And My Transcriber makes it accessible without any technical setup.

What Apple Intelligence Supports

Apple Intelligence transcription, as of early 2026, supports roughly a dozen languages:

• English
• Spanish
• French
• German
• Japanese
• Korean
• Chinese (Mandarin)
• Portuguese
• Italian
• Hindi
• Dutch
• Turkish

These are the world's most commercially significant languages. If you speak one of them, Apple's transcription works. Often well.

But there are roughly 7,000 languages spoken worldwide. Even limiting to languages with millions of speakers, Apple's list misses a huge portion of the world.

What Whisper Supports

OpenAI's Whisper model was trained on 680,000 hours of multilingual audio data. It supports transcription in 99+ languages, including:

And many more. This isn't a complete list — it's a sample. Whisper handles languages that most commercial transcription tools have never even considered supporting.

Automatic Language Detection

One of Whisper's best features is automatic language detection. You don't have to tell it what language you're speaking before you record. It figures it out from the audio.

This is important because real life isn't monolingual.

You might record a voice memo in English at work, then switch to your native language for a personal thought. Or you might work on a multilingual team where people record in whatever language is most natural to them.

With My Transcriber, you never have to configure language settings per recording. Speak whatever language you want. The model detects it, transcribes it, and records the language in the file's frontmatter.

No manual language selection. No switching settings. Just talk.

Multilingual Teams

Picture a team of five people. One speaks Swedish natively, another speaks Finnish, a third speaks Portuguese, and two speak English. They all use My Transcriber pointed at the same shared folder.

Each person records voice memos in their most natural language. The transcriptions land in the shared folder, each with the detected language in the frontmatter.

Now the team has a searchable archive of everyone's spoken notes. Need to find what the Finnish-speaking team member said about the deployment last week? Search the folder. The text is there.

Translation is a separate step — and AI tools like Claude or ChatGPT handle it well. But you can't translate audio. You can only translate text. Getting the text is the first step, and it needs to work in every language your team speaks.

Immigrants and Expats

Here's a use case that doesn't get enough attention.

If you've moved to a country where you don't speak the primary language fluently, voice memos in your native language are often the fastest way to capture thoughts. Typing in a second language is slow. Speaking in your first language is instant.

But most transcription tools assume you speak English. Or maybe Spanish or French. If your native language is Tamil, or Georgian, or Hungarian, you've been out of luck.

With Whisper and My Transcriber, you record in your native language and get accurate text back. Your thoughts are preserved in the language you thought them in. That matters.

Language Learners

If you're learning a language, recording yourself speaking and then reading the transcription is a powerful feedback tool.

Record yourself speaking the language you're learning. Read the transcription. See where Whisper misunderstood you — those are likely the parts where your pronunciation needs work.

This works for any of the 99+ languages Whisper supports. You're not limited to the major languages that language learning apps typically cover.

And because the transcriptions are markdown files, you can annotate them, add corrections, and build a personal log of your language learning progress.

Accuracy Varies by Language

I want to be honest: Whisper's accuracy isn't uniform across all 99+ languages.

For languages with lots of training data — English, Spanish, German, French, Japanese — the word error rate is very low. The transcriptions are excellent.

For languages with less training data — smaller languages, regional dialects, languages with fewer digital resources — accuracy is lower. The transcription will capture the gist but might miss words or misinterpret phrases.

This is a property of the model, not of My Transcriber. As OpenAI and the open-source community release better models, accuracy for underserved languages improves. And because My Transcriber lets you choose the model size, you can trade speed for accuracy when working in a language that needs more processing power.

Even imperfect transcription is usually better than no transcription. A rough text version of a voice memo is still searchable and skimmable — which is more than you get from untranscribed audio.

Model Sizes and Language Quality

My Transcriber lets you choose from several Whisper model sizes. The model size affects both speed and accuracy, especially for non-English languages.

• Tiny — Fastest. Good for English and major European languages in clean audio. Struggles with less common languages and noisy environments.
• Base — Default. Solid accuracy across most languages. The best balance for daily use.
• Small — Noticeably better for non-English languages. Worth the extra processing time if accuracy matters.
• Medium — Strong accuracy even for challenging languages and accented speech.
• Large-v3-turbo — The best accuracy available. Handles everything well, including minority languages and noisy recordings.

The free tier gives you tiny and base. Pro ($29 one-time) unlocks all model sizes. If you regularly record in a language that benefits from a larger model, Pro pays for itself quickly.

Runs Locally with Metal GPU

All of this runs on your Mac. No cloud API. No internet connection needed after the initial model download. No per-minute pricing.

On Apple Silicon Macs, My Transcriber uses Metal GPU acceleration to run Whisper. This means even the larger models process audio at practical speeds. A 10-minute recording in Finnish or Thai transcribes in under a minute on most Apple Silicon hardware.

Cloud-based transcription services charge per minute of audio. If you record frequently in multiple languages, those costs add up fast. Local processing means unlimited transcriptions at zero marginal cost.

It also means your audio never leaves your machine. For sensitive recordings in any language, local processing is the only option that guarantees privacy.

Why Not Just Use Apple Intelligence?

If you speak one of Apple Intelligence's supported languages and your hardware supports it, Apple's built-in transcription is convenient. It's right there in Voice Memos with no additional software.

But Apple's approach has structural limitations:

• Limited language list. About a dozen languages, focused on the largest commercial markets.
• No export. The transcription stays in the Voice Memos app. You can't get it out as a text file automatically.
• No automatic detection across many languages. When dealing with multiple languages, Apple's system is less flexible.
• Hardware gated. Requires specific Apple Silicon chips with enough Neural Engine capacity.

Apple Intelligence is fine for casual English transcription where you just want to peek at what a memo says. It's not a solution for multilingual professionals, teams, or anyone who needs the text in a file.

Real-World Examples

A Swedish developer working remotely: Records technical thoughts in Swedish while walking. The transcriptions land in the team's shared folder. English-speaking colleagues can run them through a translator if needed, but the developer captured the idea in the language it occurred to them in.

A therapist in Tokyo: Records session notes in Japanese between appointments. Each note becomes a searchable markdown file. At the end of the week, they can review what was discussed across all sessions by searching the text files.

An international family: Grandparents record stories in their native language. The recordings are transcribed and preserved as text. The grandchildren can read them, translate them, or use AI to interact with them years from now.

A graduate student researching in Arabic: Records field notes in Arabic. The transcriptions are organized by date in their research folder. They can search across months of field work for specific terms or topics.

The Frontmatter Records the Language

Every transcription file includes the detected language in its YAML frontmatter:

---
captured_at: "2026-03-22T16:45:00+02:00"
duration: "1m 58s"
language: "sv"
source: voice_memo
---

Jag tänkte att vi borde flytta deadline till nästa
vecka. Designgranskningen är inte klar ännu.

The language code follows the ISO 639-1 standard. This makes it easy to filter or sort transcriptions by language — useful when you record in multiple languages.

In Obsidian, you could create a dataview query that shows all your Swedish voice memos, or all memos in a specific language from a specific date range.

The Future of Multilingual Transcription

Whisper was a major step forward, but the field is moving fast. Newer models improve accuracy for underserved languages with each release.

My Transcriber is designed to swap in better models as they become available. When a superior multilingual model drops, it can be integrated without changing your workflow. Your output folder stays the same. Your files stay the same. Just the accuracy improves.

Apple will likely expand their language list over time too. But they'll always be limited by commercial priorities. A model trained on 680,000 hours of multilingual data will always support more languages than a model trained to serve a dozen key markets.

Summary

If you only speak English, the language advantage matters less. Both Apple Intelligence and Whisper handle English well.

If you speak any other language — or if you work with people who do — Whisper's 99+ language support is a meaningful difference. And with My Transcriber, that support is automatic. No configuration. No language switching. Just speak, and the text appears.

Voice is the most natural way to capture a thought. Language shouldn't be a barrier to turning that voice into text.

Try It in Your Language

Record a voice memo in any language. See the transcription in your folder.

macOS 15+ required. Apple Silicon only. 99+ languages with automatic detection. No configuration needed.