How Whispp and Qualcomm Brought Voice Reconstruction AI to Snapdragon X Elite

The project brought Whispp’s audio-to-audio Voice Reconstruction AI together with the on-device processing capabilities of Snapdragon X Elite. Thirty participants across nine countries used the technology during everyday calls and online conversations, providing practical feedback on performance, usability, privacy and communication quality.

Download the white paper

Bringing real-time Voice Reconstruction AI onto the device

Live voice communication leaves little room for delay, unstable processing or interruptions. For Voice Reconstruction AI to feel natural in conversation, the system needs to process audio continuously and respond almost immediately.

Whispp reconstructs whispered or affected speech into a clear, natural voice without first converting it into text. By keeping the process audio-to-audio and running it locally, the technology is designed to preserve conversational timing, expression and elements of the speaker’s identity.

For the pilot, Whispp was optimized to run on the Snapdragon X Elite platform, using its dedicated Neural Processing Unit to support real-time Voice Reconstruction AI inference.

Why the NPU matters

Running Voice Reconstruction AI on a dedicated NPU can improve both performance stability and energy efficiency. This is especially relevant for technology that needs to operate continuously during calls rather than process a single prompt or isolated task.

Testing conducted during the pilot showed that Whispp maintained processing latency of approximately 48 to 50 milliseconds when running on the NPU. CPU-based execution showed greater variation, including latency spikes and disruptions during sustained processing.

The NPU also delivered substantially longer battery runtime in the test setup. Whispp ran for approximately 310 minutes on the NPU, compared with around 130 minutes during CPU-based processing under the same conditions.

“This collaboration with Qualcomm allowed us to test Whispp where the technology ultimately needs to perform: in real conversations, on the devices people already use. Running Voice Reconstruction AI efficiently on Snapdragon X Elite shows what becomes possible when this technology moves directly onto the device.”

Joris Castermans, Founder and CEO of Whispp

Tested beyond a controlled demonstration

The pilot was designed to examine more than technical benchmarks. Participants used Whispp in professional, social and personal communication settings over a period of six weeks.

The group represented 13 different voice conditions and 10 languages. Weekly surveys, online sessions and ongoing technical support provided structured feedback on how the technology performed in everyday use.

This real-world setting helped evaluate whether on-device Voice Reconstruction AI could remain reliable, responsive and easy to use across different users, languages and communication situations.

“Thank you for the opportunity to participate and rediscover what it is like to be part of a conversation, instead of sitting back or trying to type a response and missing the moment because the conversation moved on.”

Letitia, pilot participant

What the pilot found

The results indicated strong engagement and positive feedback from participants.

All participants used Whispp at least once per week, while one-third used it daily. Seventy-nine percent rated the software as easy or very easy to use across different tasks.

Eighty-four percent gave a score of 8 or higher out of 10 when asked how likely they were to recommend the technology to someone with a voice condition. Eighty-three percent also said that on-device processing and data privacy were important factors in their experience.

A practical example of on-device Voice Reconstruction AI

The collaboration demonstrates how dedicated on-device hardware can support a demanding, continuous Voice Reconstruction AI workload directly on a personal device.

For Whispp, this means bringing Voice Reconstruction AI closer to the devices and communication platforms people already use. For users, it means lower reliance on cloud connectivity, more consistent real-time performance and voice data that can remain on the device.

The pilot also provides a practical foundation for the further development of real-time Voice Reconstruction AI experiences across mobile and PC platforms.

Read the full white paper

The complete white paper covers the pilot methodology, participant experiences, technical performance and lessons from deploying real-time Voice Reconstruction AI on Snapdragon X Elite.

It includes:

• CPU and NPU latency comparisons
• Battery-consumption results
• An explanation of Whispp’s audio-to-audio Voice Reconstruction AI
• Usability and adoption findings
• Participant experiences and case studies
• The role of on-device processing in real-time communication

Download the white paper

Whispp’s assistive applications are under development and are not currently offered as medical devices.