Product news

We have launched the first version of our language detection feature to simplify your workflow when you aren't sure which language is being spoken.

How it works:

You’ll now find Detect language placed inside the target language picker. Record a few seconds of speech, and the app will analyze the audio to suggest the most likely languages. To keep the experience simple, we currently present up to the top three results.

Powered by Whisper:

The feature leverages a self-hosted Whisper model to handle identification. Because this is a first version, we are currently using the model's native performance and language support to learn how to best optimize the feature for you.

Confidence scores:

Each suggested language is accompanied by a colored pill showing a percentage (e.g., 90%).

• What it means: This is a "confidence score" determined by the model.
• Smart filtering: To ensure you only see relevant options, we automatically filter out any languages with a confidence score below 0.5%.