Speech language: Help Recognition Follow the Language You Speak

A short recording can give a recognizer less context than a long conversation. If you usually dictate in one language, explicitly selecting it can help the speech service interpret the recording in the intended language. Sonicribe puts this choice in Settings → Listening → Speech language.

This setting describes the speech you are providing. It does not request a translation or determine the language of the app's buttons. Those are separate decisions, so start by fixing recognition before asking a formatting style to transform the output.

Choose a useful recognition setting

• Open Settings, find Spoken language, and open Speech language.
• Choose Auto-detect when you want the recognizer to infer the spoken language.
• If you predominantly speak one language, select that language from the available list.
• Record a short sentence containing the names or terms you commonly use.
• Compare the recognized wording with what you said, then keep the setting that works for your task.

Choose Auto-detect or a spoken language such as English, Korean, Japanese, Chinese, Spanish, French, German, Portuguese, Italian, Russian, Arabic, Hindi, Thai, or Vietnamese. Use a short sentence with names and numbers to compare the recognition setup you plan to use.

Auto-detect is the global default. A style's Voice → Language can override the global choice, so check that field if one style behaves differently from another.

Test a real mixed-language sentence

If you write Korean sentences with English product names, test that combination rather than a sentence containing only common words. For example, you might dictate a Korean project update that includes “Sonicribe” and an English feature name. Check whether the result keeps the intended spelling and surrounding language.

Do not assume that locking one language will solve every mixed-language recording. Try a short comparison with Auto-detect, and use Words for recurring names and specialized terms. Speech hints depend on the provider; the app's hint notice explains that MAI-Transcribe-1.5 does not support speech hints.

Request translation separately

If you need an English draft from speech in another language, describe that transformation in a formatted style's custom instructions, such as “Translate into English and preserve every date and qualification.” The formatting model then has a writing instruction to follow. Exactly as spoken does not run that formatting stage.

Always review translated names, quantities and commitments. A polished translation can still lose a nuance, and selecting English under Speech language is not a substitute for a translation rule.

If recognition unexpectedly changes after a setting adjustment, return to Auto-detect and repeat a small test before changing additional settings. Also check the microphone and selected speech model; language selection cannot recover audio that was not captured clearly.

See the Speech language instructions, then use the independent model guide to keep recognition and formatting decisions separate.