Technology
Real-Time On-Device Voice Reconstruction AI
Whispp’s Proprietary Voice AI
At Whispp, we’re pioneering on-device audio-to-audio AI, a novel technology that transforms how people communicate, whether they’re whispering, speaking in noisy environments, or living with a voice condition.
Traditional tools
Traditional voice processing tools like noise suppression, spectral masking, or voice isolation are good at removing unwanted sound but they can’t restore what’s missing. Whispp goes a step further. Using Whispp’s proprietary voice AI, we don’t just clean audio, we recreate lost or degraded speech information, delivering in real-time natural, expressive and intelligible communication where it wasn’t possible before.
Whispp’s advanced voice AI reconstructs and restores speech in real time, reviving lost pitch, tone, and natural intonation.
Voiced speech
Figure a shows the spectrogram of regular voiced speech. Key features of voiced speech are noted below:
- Fundamental frequency (F₀): visible as evenly spaced horizontal striations representing vocal fold vibrations (pitch).
- Harmonic structure: multiple frequency bands stacked above the F₀, giving richness and timbre to the voice.
- Formants (F₁, F₂, F₃, …): darker bands that correspond to resonances in the vocal tract, shaping vowel sounds.
- Energy concentration: clear patterns of energy that vary dynamically across time and frequency, reflecting natural rhythm and prosody
Whispered speech
Figure b shows the spectrogram of whispered speech. You can see the loss of:
- Fundamental frequency (F₀): no visible pitch band because the vocal folds do not vibrate.
- Harmonic structure: replaced by diffuse, noisy energy.
- Natural intonation: speech sounds flat and monotone since there’s no pitch variation.
Whispered speech retains some formant information, allowing words to remain intelligible, but loses the voicing cues that make speech sound natural and expressive.
Whispp Reconstructed Speech
Figure c shows the spectrogram of reconstructed speech using Whispp’s Proprietary Voice AI. Whispp restores the key features of natural spoken speech from a whispered sample:
- Reconstructed fundamental frequency and harmonics, bringing back natural pitch and tone.
- Enhanced formant clarity, improving intelligibility and timbral accuracy.
- Reintroduced prosody and emotional expressiveness, allowing speech to sound authentically human again.
(a) Spectogram of regular voiced speech
(b) Spectogram of whispered speech
(c) Spectogram of Whispp reconstructed speech
Figure 1. (a) Spectogram of regular voiced speech, (b) spectorgram of whispered speech, and (c) spectogram of Whispp reconstructed speech. © Whispp.
Figure 1. Spectograms of various speech forms showcasing Whispp reconstruction AI technology. © Whispp.
Personalizing voices
By providing recordings, your Whispp voice will sound like your own healthy voice!
In the Whispp app you can use your Personal Whispp voice for your video or audio calls and messages. Stay connected with family, friends, and others in a way that feels familiar and comfortable.
Whispp reconstructs the key features of natural spoken speech using its proprietary voice AI technology. It does this on-device, in real-time and in any language.
Whispp’s technology can also be applied to affected voices, or heavily noisy environments. The result is speech that not only sounds clear but feels authentic, expressive, and true to the speaker’s original voice.
