VivaDicta - Speech to Text

Your words, ready the moment you stop talking. VivaDicta 3.8.0 adds real-time transcription, finally records through your headphones, and refreshes the transcription and AI model lineup. Soniox Realtime A new transcription option that sends your speech to Soniox while you talk, instead of uploading the recording after you stop. The text is ready almost the moment you finish. Pick "Soniox Realtime" in the model list to use it. If the connection drops it quietly falls back to the normal upload, so nothing is lost. The standard Soniox model also returns results with noticeably less waiting. Your Headphones' Microphone If you're wearing AirPods or a Bluetooth headset, VivaDicta now records through them instead of falling back to the iPhone microphone - the same way Phone and Voice Memos behave. A new Microphone setting lets you pin the iPhone microphone instead, which keeps your headphones in full-quality audio while you dictate. On iOS 26, recording uses Apple's high-quality Bluetooth mode where supported, so music and podcasts no longer drop to muffled headset sound. Sign In With Grok Connect a SuperGrok or X Premium account and use Grok for AI processing without an API key. Sign-in happens in the browser with a short code, the same way you'd connect a TV app. xAI limits which plans can connect an app, so an API key is still there as a fallback and unlocks the full model list. Smarter Filler Removal Hesitation sounds like "uh" and "um" are now removed based on the transcript's language, so real words are no longer stripped by mistake. German "um" (as in "um zu") and Portuguese "um" stay intact. A new Remove Filler Words toggle in Settings > Advanced turns the cleanup off entirely and keeps every word exactly as spoken. New Transcription Models AssemblyAI Universal-3.5 Pro is now the default AssemblyAI model, with native code switching for multilingual speech and better speaker labels. Soniox moves to v5 for higher accuracy across 60+ languages. Also new: Gladia Solaria-3 for English, French, German, Spanish, and Italian, Speechmatics Melia-1 which detects and switches between 56+ languages on its own, OpenAI GPT-4o Transcribe Diarize for speaker-labeled transcripts, and full-size Whisper Large V3 on Groq. Latest AI Models Claude Sonnet 5, the GPT-5.6 family, Grok 4.5, and GLM-5.2 join the lineup, and Kimi and Cerebras get fresher defaults. The GPT-5.6 models are available on the OpenAI sign-in path too, so you get them without an API key. Retired models are upgraded to their replacements automatically, so no mode breaks and nothing needs re-picking. Quality of Life The keyboard session timeout gains a Never option, hot mic sessions get an in-app stop control, and adding your own transcription server now lets you choose how audio is sent, so endpoints that only accept it inside a JSON request finally work.