Multilingual meeting recording and Whisper
Global engineering teams and international client calls require flexible language support. Oyma incorporates OpenAI’s multilingual Whisper speech models, executing completely on-device across 13 core languages.
Supported languages
The on-device speech engine supports the following languages with high transcription fidelity:
| Language | Language Code | Model Recommendation |
|---|---|---|
| English | en | Base (148 MB) or Small (488 MB) |
| Spanish | es | Small (488 MB) or Large v3 Turbo (1.6 GB) |
| French | fr | Small (488 MB) or Large v3 Turbo |
| German | de | Small (488 MB) or Large v3 Turbo |
| Italian | it | Small (488 MB) or Large v3 Turbo |
| Portuguese | pt | Small (488 MB) or Large v3 Turbo |
| Dutch | nl | Small (488 MB) or Large v3 Turbo |
| Japanese | ja | Large v3 Turbo (1.6 GB) |
| Chinese (Mandarin) | zh | Large v3 Turbo (1.6 GB) |
| Russian | ru | Small (488 MB) or Large v3 Turbo |
| Korean | ko | Large v3 Turbo (1.6 GB) |
| Polish | pl | Small (488 MB) or Large v3 Turbo |
| Swedish | sv | Small (488 MB) or Large v3 Turbo |
Language detection modes
Configure your language strategy in Settings › Transcription › Language:
- Auto-Detect (Default): Whisper evaluates the first speech segments and automatically locks onto the spoken language.
- Forced Language: Select a specific language manually (e.g. German) to eliminate detection ambiguity on short calls or accented speech.
Bilingual conversations
In mixed-language meetings (e.g. an engineering call alternating between Spanish and English), the Whisper Large v3 Turbo model delivers optimal accuracy. Its expanded parameter space switches between phonetic alphabets seamlessly without dropping sentences.
Related guides
- Model sizing in Transcription models.
- Audio transcription in Transcribing audio and video files.
- Meeting recording in Record a meeting.
Questions
Which languages does Oyma transcribe?
Oyma supports English, Spanish, French, German, Italian, Portuguese, Dutch, Japanese, Chinese, Russian, Korean, Polish, and Swedish.
Does language auto-detection introduce latency?
Whisper analyzes the initial 30 seconds of speech to determine the dominant language with minimal latency overhead.
Can meeting summaries translate into English?
Yes. You can instruct your local Ollama or BYOK model to summarize foreign language transcripts directly in English.
