Speech and Formatting Models: Choose Each Stage for Its Own Job

When a dictated draft goes wrong, the first useful question is where the problem began. Did the app recognize the wrong word, or did the writing stage reshape a correct transcript in an unwanted way? Sonicribe separates Speech transcription from Text formatting so you can choose and troubleshoot those jobs independently.

The speech model processes your recording. A text formatting model, when the style uses one, works on the recognized text. Choosing a different speech model does not automatically mean you have selected a different writing style.

Check the two selections in your style

• Open Styles and the style's Rewrite.
• Open Speech transcription and choose a speech model currently available to your account.
• Open Text formatting and choose an available formatting model for a formatted style.
• For direct transcription, select Exactly as spoken, which uses No formatting.
• Select the style and use Today's Start listening for a short example representative of your task. Review names, numbers, meaning and layout separately.

For global shortcut recording into another app, Sonicribe chooses a linked app/site style or Styles → Default style for an unmatched destination. Configure that selection too; merely selecting a style card does not guarantee that global recording will use its models.

The Models view also shows the model catalog. Availability is part of the service state: a visible card or remembered selection does not guarantee that a provider is currently ready. Use selectable models and read any disabled reason or account warning shown in the app.

Diagnose with a controlled example

Suppose you dictate: “Send Mira the draft after the legal review, not before.” If the recognized name is wrong, check speech language, microphone quality and vocabulary. If the name is right but the output says the draft will be sent immediately, review the style's instructions and the formatted result.

Use a short, repeatable phrase when comparing settings. Change one selection at a time and keep the task similar. Your own correction effort is more useful than assuming that a model name guarantees a particular accuracy or speed.

Sonicribe does not provide a guarantee that every catalog entry will remain available. Speech and optional formatting also use the same account credit allowance. Skipping formatting removes that processing stage, but it does not make speech transcription offline or free of credit use.

Recover from an unavailable selection

If recording is blocked because a model is unavailable, choose another model that the app marks selectable. If none are available, check Account, the connection and any service warning. Repeatedly changing writing rules will not resolve a provider readiness or credit problem.

Keep your model choices with each style. Open Voice → Speech transcription for recognition and Rewrite → Text formatting for the writing format. Compare a short sample with your preferred choices, then use Models to review the listed services and credit rates.

Use the model instructions for the control reference. The Exactly as spoken article explains how to establish a baseline before tuning formatting, and custom styles explain how to describe the output you want.