Transcription in 99 languages — with a real DACH focus
Auto-detect, mid-sentence code-switching and dialect handling — all on one engine.
3 free transcriptions · no credit card · data stays in Germany
99+ Sprachen · Auto-Detect
DeepScript runs a Whisper-compatible engine and supports 99 languages — from German and English through French, Italian, Dutch, Spanish, Polish, Portuguese, Russian, Mandarin, Japanese, Korean, Arabic, Hebrew and Hindi to Swahili, Norwegian, Swedish, Finnish, Greek, Turkish and Czech. Auto-detect typically identifies the language within the first 5–10 seconds of audio. Mid-sentence code-switching — say a German-English business meeting — is transcribed correctly. DACH languages additionally benefit from our specialised fine-tuning; smaller languages perform closer to the generic Whisper level.
Proof
Why we can claim this
99 languages via the Whisper base
Full Whisper language coverage. Per-language accuracy depends on training share — large languages are excellent, exotic ones are good.
Auto-detect within 5–10 seconds
You don't have to specify the language. The model listens briefly and decides — on multilingual recordings, it re-decides mid-stream.
Native code-switching
"Lass uns das im Q3-Review aligned bekommen, ja?" — mixed German-English sentences are transcribed correctly without language-switch glitches.
DACH dialects fine-tuned
Swiss German, Bavarian, Viennese, Low German — where generic Whisper trips, our Premium model keeps going.
In practice
What this looks like in practice
Auto-detect, mid-sentence code-switching and dialect handling — all on one engine.
- Top-tier quality (≤ 5% WER on clean recordings) for German, English, Spanish, French and Italian.
- Solid quality for Dutch, Portuguese, Polish, Russian, Mandarin, Japanese and Arabic.
- Smaller languages (Hindi, Swahili, Tamil, Welsh …) are available; WER can be noticeably higher depending on audio quality.
- Detected language is stored on the transcription (`detectedLanguage`); you can override it after the fact.
- Translation is no longer built in directly — if you need the transcript in another language, use MCP with an AI client (Claude, ChatGPT).
99+ Sprachen · Auto-Detect
How to use it
Up and running in a few steps
- 1
1. Pick a language or auto-detect
In the upload form: a dropdown with 99 languages plus "Auto-detect" as the default. If you know the language, setting it explicitly is marginally more accurate.
- 2
2. Upload — auto-detect runs at the start
The engine listens to the opening seconds and decides which decoder to use. On multilingual recordings, it switches seamlessly.
- 3
3. Verify detected language
The detected language is shown in the result header. If it's wrong — e.g. you spoke Swiss dialect and the model picked Standard German — you can re-transcribe with an explicit language choice.
- 4
4. Export or process further
Original language → TXT/SRT/VTT/JSON. For translation into other languages: hand the transcript to your AI client (Claude, ChatGPT) — via MCP or copy & paste.
FAQ
Frequently asked questions
Which languages are supported exactly?+
The full list of all 99 languages is available via `GET /v1/languages`. The most important ones: German, English, French, Italian, Dutch, Spanish, Polish, Portuguese, Russian, Mandarin, Japanese, Korean, Arabic, Turkish, Hindi, Swedish, Norwegian, Finnish, Greek, Czech and many more.
How accurate is Hindi or Arabic compared to German?+
On clean recordings, Hindi and Arabic typically land at 8–12% WER, German at 3–5% (Premium) or 5–8% (Standard). The gap comes from the Whisper base training having more hours of European languages. Still usable for notes, prep and downstream machine processing.
Auto-detect — how fast is it?+
Language identification runs in the first 5–10 seconds of audio. You won't notice a perceptible delay — the overall process (upload → transcription) starts immediately after upload.
Code-switching — does it really work?+
Yes, especially for the most common pairs (German ↔ English, Spanish ↔ English). On "Lass uns das ASAP bouncen" the model recognises both languages and transcribes them natively. Exotic mixes (Mandarin ↔ French) work but are less reliable.
Can I automatically translate the transcript?+
Not directly from us — we removed the server-side AI endpoints (summarize, translate) in May 2026. Instead, hand the transcript to your AI client (Claude, ChatGPT, …) — most elegantly via MCP, so the client reads your transcriptions directly.
See it for yourself
Upload a file and see the result in minutes. Three transcriptions free, no credit card.