Transcribe Audio in Any Language (60+)

Apple Transcribes 10 Languages. Your Audio Speaks Many More.

Apple's built-in transcription covers exactly ten languages: English, Spanish, Portuguese, Italian, French, German, Japanese, Korean, and Simplified and Traditional Chinese. And even inside that short list there is no import, no translation, no summary. Outside it — Arabic, Hindi, Thai, Vietnamese, Cantonese, Russian, Dutch, Swedish, Turkish, Polish, Indonesian, Ukrainian — there is no Apple transcription option at all.

Sound files don't care about language walls. The voice memo from your São Paulo supplier arrives in Portuguese. The visiting professor's lecture arrives in Japanese. The family message from your aunt in Hong Kong arrives in Cantonese. The recording is in the speaker's language, not your settings.

This page shows how to get a searchable, copyable, translatable transcript from audio in 60+ languages on one iPhone app — with honest accuracy expectations per language. Transcribe foreign language audio on your iPhone, keep the transcript in the source language, and read the translation beside it.

How to Transcribe Audio in Any Language

Step 1 — Record or import. Tap record in the app, or import a file. Audio imports cover m4a, mp3, wav, caf, aac, aiff, and mp4a; video imports (mp4, mov, m4v, avi, mkv) have the audio track extracted automatically. You can also hand files to the app through the iOS share sheet from WhatsApp, Zoom, Voice Memos, and anywhere else audio can be shared.

Step 2 — Whisper transcribes, server-side. Your audio is processed by openai/whisper-large-v3-turbo, with large-v3 as fallback. Whisper auto-detects the language, and it's usually right. When it isn't — or when you already know the language — set it explicitly in the picker: 100 entries, the 99 Whisper languages plus Cantonese, with a popular row that surfaces English, 中文, 廣東話, 日本語, 한국어, and Español.

Step 3 — Read, translate, summarize. The transcript stays in the source language. When the detected language differs from your app's preferred language, the transcript is automatically translated and summarized into your preferred language — no toggle, it's built in.

Honest notes: the app runs on iPhone and iPad only, and transcription requires internet. Free gives you 60 minutes of lifetime transcription — one-time, not monthly. Pro is $5.99/month for 10,000 minutes per month, with a 90-minute cap per recording (about 50MB per file). Notes shared by link are protected with AES-256-GCM zero-knowledge encryption: the key rides in the URL hash, and links expire after 24 hours.

The 12 Languages That Matter Most

Twelve languages people ask about most, compared against Apple's built-in option and Whisper's actual accuracy tiers.

Transcribe Each Language

Spanish sits at near-English accuracy with a 3-6% word error rate, and Whisper handles both LatAm and Castilian accents. Clean enough for lectures and interviews, and the in-app translation reads naturally in English. Start with how to transcribe Spanish audio.

French transcribes at a 4-8% word error rate, and Whisper copes with Québécois accents that trip up most tools. Podcasts, client calls, and French media come out readable; when French isn't your first language, the auto-translation does the heavy lifting. See how to transcribe French audio.

German lands in the near-English tier at 4-7% word error — impressive for a language that builds compounds like Donaudampfschifffahrtsgesellschaft. Lectures, business audio, and long podcasts transcribe cleanly. Read more about transcribing German audio.

Japanese runs at a 5-10% character error rate — better than dedicated CJK models, by Whisper's own benchmarks. University lectures and meetings transcribe well enough to follow, with only the odd Kanji swap. See how to transcribe Japanese audio.

Portuguese transcribes at 4-7% word error with a Brazilian bias, which suits sales calls and meetings out of São Paulo. European Portuguese works, just slightly less consistently. Learn how to transcribe Portuguese audio.

Korean transcribes at a 5-8% character error rate, with strong results on interviews, lectures, and K-content. The transcript stays in Hangul, and the English translation preserves sentence structure well. Details in our Korean transcription guide.

Mandarin runs at 10-14% character error — good, not perfect. Business calls and podcast subtitles are the sweet spot; numbers and proper nouns deserve a quick scan. Read how to transcribe Mandarin audio.

Cantonese is the weakest of the majors at roughly 13% word error — but Whisper is still the best open model for it. Set 廣東話 explicitly in the picker and expect draft-quality output: fine for HK diaspora family messages and media. See our Cantonese guide.

Arabic transcribes well on Modern Standard Arabic and weaker on dialects, so Gulf business audio, broadcast media, and interviews are the sweet spot. Diacritics are dropped, standard for Arabic transcripts. Learn more about transcribing Arabic audio.

Hindi is workable at 9-14% word error, and setting Hindi explicitly in the picker matters more than for most languages — especially with Hinglish code-switching. Devanagari notes and lectures come out readable. See how to transcribe Hindi audio.

Vietnamese sits in the draft tier at 15-22% word error, but it gets the diacritics right — which is exactly what most tools get wrong. Family voice messages and diaspora audio turn into searchable text. Read our Vietnamese transcription guide.

Russian is stronger than most people assume: 5-7% word error. Research interviews, podcasts, and long-form audio transcribe cleanly — just make sure auto-detect isn't reading Ukrainian as Russian. Start with transcribing Russian audio.

Common Pitfalls (and What Actually Works)

• Auto-detect is excellent for most languages and weakest for Cantonese, Ukrainian (which it tends to misread as Russian), and code-switched Hindi/Hinglish. When the language is one of those — or when accuracy matters — set the language explicitly in the picker.
• Whisper large-v3-turbo trades a little accuracy for roughly 8x speed. Long lectures should be split into recordings under the 90-minute cap rather than one giant file.
• For transcripts you'll reuse, translate inside the app rather than pasting into a translator. The translation and summary stay attached to the original, with the source-language text intact.
• Don't expect speaker labels. The app transcribes; it does not attribute who said what. Note names in your own summary if attribution matters.
• Voice Memos dictation covers ten languages, cannot import files, and cannot translate. For anything beyond that you need a real transcription app — which is the point of this one.

Frequently Asked Questions

Which languages does Audio Notes AI transcribe?

The language picker has 100 entries: the 99 Whisper languages plus Cantonese — 60+ spoken at native level. If your language isn't in the picker, it isn't in the models; that's the complete list.

Can I transcribe a language Apple's Voice Memos doesn't support?

Yes. Arabic, Hindi, Thai, Vietnamese, Cantonese, Russian, Dutch, Swedish, Turkish, Polish, Indonesian, Ukrainian, and dozens more all transcribe in the app. Whisper auto-detects the language, or you set it explicitly in the picker.

Can I translate a transcript to English?

Yes — automatically. When the detected language differs from your app's preferred language, the transcript is translated and summarized into your preferred language. The original transcript stays in the source language, untouched.

Is transcription on-device or server-side?

Server-side. Audio is uploaded to openai/whisper-large-v3-turbo (with large-v3 fallback) and transcribed in the cloud, which is why transcription requires an internet connection.

How much does it cost?

Free includes 60 minutes of lifetime transcription — a one-time allowance, not monthly. Pro is $5.99/month for 10,000 minutes per month, capped at 90 minutes per recording.