Quiet dictation, explained

Quiet voice dictation on Windows without simply turning everything up

Written and technically reviewed by the Detto product team · Updated 24 July 2026

Short answer: Detto analyzes the microphone level over the whole capture, estimates the background floor, identifies voice-like activity, and only then applies a safe amount of gain. You can choose Auto, Quiet voice, or Noisy room in Settings.

Why whispering is harder than normal dictation

A fixed microphone threshold creates an unpleasant trade-off. Set it too high and a soft sentence is discarded as silence. Set it too low and a fan, laptop hum, or idle microphone noise can look like speech. Blind amplification is not enough either: it raises background noise with the voice and can push loud peaks into clipping.

Detto beta.18 replaces its old single cutoff with a lightweight local signal tracker. It considers the low end of the measured level distribution, peak energy, the portion of active frames, and the selected sensitivity profile. This is deliberately smaller and easier to validate than adding a new neural noise-suppression dependency before launch.

What happens after you speak softly

1. Local signal analysis: the app measures frame energy and peak headroom while recording. It does not inspect the meaning of the sentence at this stage.
2. Conservative voice decision: digital silence is kept separate from low-confidence background sound and voice-like activity.
3. Bounded gain: weak speech is amplified inside the existing WAV encoder. The selected profile sets a maximum, and the measured sample peak adds a second limit to prevent PCM clipping.
4. Normal transcription: the resulting audio goes to the speech provider the user already configured. A quiet capture that returns no usable words may try the other already-configured provider once.

Choose the right sensitivity profile

Start with Auto. In Detto, open Settings → Voice sensitivity. Choose Quiet voice only when soft speech is your normal pattern; choose Noisy room when background sound is the actual problem.

Privacy, cost, and realistic limits

The new detector and gain step run locally in the desktop renderer. They do not add a service, model, subscription, or stored audio copy. Cloud transcription still follows the provider configuration chosen by the user. If both speech providers are configured, only a quiet capture that the first one cannot recognize can generate one additional transcription request.

Quiet mode improves the signal Detto sends; it cannot recover words a microphone never captured. Distance, microphone direction, automatic gain control in the driver, and competing voices still matter. For the best result, keep the microphone within its intended range and confirm that the level meter moves during setup.

Frequently asked questions

Can Detto transcribe a whisper?

It is designed to recover substantially softer speech than the previous fixed threshold, especially with Quiet voice selected. Recognition still depends on hardware, distance, room noise, and the speech provider, so “every whisper” would be an unrealistic promise.

Does the mode change my written style?

No. Voice sensitivity affects capture and pre-encoding gain. Auto, Messages, Email, Prompt Engineer, Natural, and Verbatim continue to control how recognized words are written.

Will normal speech become distorted?

No gain is added when the measured level is already healthy. When gain is needed, measured headroom caps it before PCM conversion.