Transcript Came Out in the Wrong Language

If a Hedy transcript comes out in the wrong language, the most common cause is the meeting language setting being out of sync with what was actually spoken. The fix is in Settings > Profile > Language Preferences > Meeting/Class Language. The behavior also depends on which speech recognition provider you’re using — some auto-detect language, while the default (local Whisper) does not.

Where the Language Setting Lives

Hedy has one app-level language setting for transcription — Meeting/Class Language — that the default speech recognition provider (local Whisper) uses directly. Cloud providers (Deepgram, OpenAI) and the experimental Parakeet engine have their own per-provider language settings that may override or ignore the app-level setting.

Open Hedy’s Settings
Go to the Profile tab
Find Language Preferences
Set Meeting/Class Language to the language you’ll actually be speaking

You can also change this from the session-start screen — there’s a speech recognition language dropdown there. Note that changing it on the session-start screen persists the change to your account, so the next session will use the new language too. There is no separate “this session only” override.

If you have never explicitly picked a language, Hedy defaults to the language you chose during onboarding (suggested from your device locale), or English if nothing was set.

Why This Depends on Your Speech Recognition Provider

Hedy supports four speech recognition providers — two local, two cloud — and they handle language differently. You can see and change your active provider at Settings > Speech & AI > Speech Recognition Options.

Provider	Type	Language behavior
Local Speech Recognition (Whisper) — default	Local	Uses the Meeting/Class Language you configured. Does not auto-detect. If you set Spanish but speak German, the transcript will be wrong.
Local Speech Recognition (Parakeet) [Beta]	Local (Apple Silicon Macs and supported iPhone/iPad models)	No explicit language parameter is sent — the model identifies language from audio. Works best for English and major European languages. May misidentify similar languages (German vs. Dutch, Spanish vs. Portuguese, etc.).
Deepgram (requires your own API key)	Cloud	Defaults to Multi (auto-detect). Can also be set to a specific language.
OpenAI (requires your own API key)	Cloud	Defaults to Auto-detect language. Can also be set to “Use meeting language”.

If you’re on the default Whisper provider, the Meeting/Class Language must match what’s being spoken — this catches a lot of wrong-language reports. If you’re on Parakeet, the model picks the language for you, but can get confused between similar-sounding languages. If you’re on Deepgram or OpenAI in auto-detect mode, language is usually figured out from the audio — but you can still force a specific language in the provider’s own settings if it’s getting it wrong.

Common Scenarios

“My transcript is in English but I spoke German.” You’re likely on local Whisper (the default) with Meeting/Class Language set to English. Switch the language in Settings > Profile > Language Preferences. If you’re on Parakeet and your spoken language is close to English (German, Dutch, French), this can also happen — try Whisper with the language set explicitly, or move to Deepgram/OpenAI for auto-detect.

“I speak two languages in the same meeting and only one gets transcribed correctly.” Whisper assumes a single language for the whole session, and Parakeet has limited multilingual handling. For multilingual meetings, switch to Deepgram (with multi language setting) or OpenAI (auto-detect), both of which require your own API key.

“Place names and proper nouns are wrong, but the language is right.” That’s a vocabulary issue, not a language issue. See Fix Poor Transcription Quality and the Custom Vocabulary guide.

“I switched languages mid-session and the new language isn’t being picked up.” Behavior depends on the provider. Local Whisper (the default) re-reads the Meeting/Class Language for each transcription request, so changes mid-session can take effect on subsequent audio. Cloud providers (Deepgram, OpenAI) lock the language at the start of the WebSocket session — changes apply only to the next session. For consistent results regardless of provider, end the current session and start a new one after changing the language.

Languages Hedy Supports

The full list of meeting languages Hedy supports (on desktop):

English, Afrikaans, Arabic, Catalan, Croatian, Czech, Danish, Dutch, Finnish, French, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Malay, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovenian, Spanish, Swedish, Tagalog/Filipino, Thai, Turkish, Ukrainian, Vietnamese, Chinese Mandarin, Chinese Cantonese.

On iPhone, iPad, and Android, the list is the same except Arabic, Hindi, and Thai are not available.

Hedy’s App Language vs. Meeting Language

Two different settings, easy to confuse:

App Language — the language of Hedy’s user interface (buttons, menus, settings). Set in your Hedy app settings.
Meeting/Class Language — the language Hedy expects you to speak during sessions. This is what affects transcription.

Changing one doesn’t change the other. You can have Hedy’s UI in English and your Meeting Language in Spanish.

Fix Poor Transcription Quality — for accuracy issues even when the language is right
Speech Recognition Providers in Hedy — the full breakdown of each provider
Custom Vocabulary Guide — for proper-noun and jargon accuracy

Still having trouble? Contact us through the chat widget with the language you were speaking, the language you got in the transcript, and which speech recognition provider you’re using.