How Whispp turns a whisper into a voice without writing it down

A whisper still carries information about the words being spoken. Whispp’s voice reconstruction system works directly on that audio, rebuilding it into a voiced signal as the person speaks. It does not need to transcribe a sentence and then have a text-to-speech system read the sentence back.[1][2]

The company’s technical explanation starts with what changes in a whisper. Ordinary voiced speech has a fundamental frequency, heard as pitch, and related harmonics. Whispering loses that regular vibration, but retains patterns shaped by the mouth and vocal tract, including the resonances that distinguish vowel sounds. Whispp says its model reconstructs the missing pitched components and refines the speech’s resonance and rhythm.[1]

Reconstruction supplies missing characteristics; amplification only makes the existing signal louder. Whispp also says its model can operate across languages.[1]

Fast enough to stay in the conversation

In a July pilot with Qualcomm, Whispp ran the system on the neural processing unit, or NPU, in Snapdragon X Elite PCs. Thirty participants across nine countries tried it over six weeks, according to the company.[2]

Whispp reported processing latency of about 48–50 milliseconds on the NPU. In its battery comparison under the same test conditions, runtime was roughly 310 minutes on the NPU against 130 minutes on the CPU. These are the developer’s measurements for that setup; the latency figure describes processing, rather than the total delay across a phone or video call.[2]

A conversation is made of turns, interruptions and small opportunities to respond. An audio system that processes speech continuously can preserve a different rhythm from one that asks its user to finish typing a message. The device work matters because the transformation has to keep going while the call does, without making every reply feel like a separate submission.

The call itself is part of the product

Independent experience with Whispp’s earlier smartphone calling app offers another view. Parkinson’s UK’s Tech Guide had six people with Parkinson’s try it for four weeks in January 2026. This was a small user panel, separate from the Qualcomm pilot. Half reported louder, clearer telephone speech; others found that it did not suit them.[3]

The panel also identified friction outside the voice model. Outgoing calls appeared from an unfamiliar number, and incoming calls had to be initiated through the app. That complicated return calls from organisations such as hospitals. Some reviewers preferred their personalised voice to the presets; one found that family understood them less well through the app.[3]

Whispp’s current PC offering takes an integration approach. WhisppWAVE is a native Windows SDK to incorporate into an existing audio path, with processing on the CPU or NPU. The company describes it as working across conferencing and calling applications without a cloud dependency for reconstruction. That is an integration product, rather than a claim that every Windows meeting app already includes it.[4]

The PC product is presented as communications technology, and Whispp says it is not intended for medical use. The earlier user panel should therefore be read as experience with a particular calling app, rather than evidence of a treatment effect or a result guaranteed for the newer engine.[4][3]

For this technology to become useful in everyday conversation, the reconstructed voice and the familiar call need to meet in the same place. A clear sound is one achievement. Being able to answer, be recognised and take a turn without reorganising the conversation is the next.

Report an error or suggest a correction