v1.5.0
Speech engine upgraded to Soniox v5: more accurate mixed-language recognition; pausing mid-sentence in a long dictation no longer cuts you off.
Upgraded to Soniox v5. A new generation of the recognition model, more accurate and stable across languages, mixed Chinese/English/French, accents, and noisy environments, with more reliable language-switch detection too. The model has already switched to v5 by default — no action needed on your part.
Long sentences no longer get cut off. You can now speak for a long stretch at once — pausing mid-way to think or find the right words won't cut you off early anymore. Text only finalizes and pastes once you release the hotkey. Speaking a whole paragraph now feels a lot more relaxed.
Also, the model picker was simplified down to pure real-time streaming, the settings UI got a refresh, and several stability issues were fixed (occasional stutter on hotkey press, audio device switching).