Recording is not the same as transcription
Recording saves audio. Transcription attempts to recognize speech and write text. A summary is another step that works from that text. In VoxTrail, saving a recording or receiving one from Apple Watch does not silently start the other steps.
Open a saved recording on iPhone, choose local transcription, select its spoken language and start processing. Check the result against the original, especially names, amounts and decisions. Speech recognition can make mistakes.
What does on-device transcription mean here?
The current app requires Apple Speech to perform local recognition on the iPhone. That route is separate from the optional cloud route. If the chosen language or device cannot run the local task, the app reports the problem; it does not automatically fall back to an online recognizer.
Recording and local storage do not need a cloud account. Recognition availability still depends on the device, language and system support, so this is not a claim that every language works offline on every iPhone.
When would you choose cloud speech to text?
You may choose to configure a compatible cloud transcription provider using your own endpoint and API key. That route sends original audio to that provider after confirmation. The provider’s account permissions, regional availability, billing and data policy apply.
VoxTrail currently supports an optional Qwen-ASR route. It does not include official cloud credits. The current output aligns text to audio windows rather than providing word-by-word timestamps.
Are transcription languages the same as website languages?
No. This website is available in English, Simplified Chinese, Traditional Chinese, Japanese and Spanish. Those are website translations. They do not establish the app’s interface languages or prove that a specific recognition language is available on a particular device.
Choose the language actually spoken in the recording. Transcription writes recognized speech; it is not a translation feature.
Does local AI have the same requirements?
No. On-device summaries and to-dos use a different capability from speech recognition. They additionally require iOS 26, an Apple Intelligence-compatible device and a ready model in an eligible region. You request summaries manually after obtaining text.
Without a usable AI model, original audio, saved text and basic search remain separate features. See the product facts for current development and launch status.