EIC Accelerator Winner Whispp and the On-Device Voice AI Race

Whispp's idea rests on an observation from stuttering research: people who stutter severely often speak fluently when they whisper. Whispp, a Leiden company founded by Joris Castermans, who stutters himself, built an AI that takes the whisper and turns it back into a clear voice, ideally the speaker's own, in real time and on the phone or laptop itself. The EIC Accelerator selected the project at its October 2025 cut-off for blended finance, and CORDIS records an EU contribution of €2,498,319 for a project running from 1 June 2026 to 31 May 2028.

The project's title calls it "assistive AI technology for human connection". What Whispp sells in November 2026 is something broader: software kits that let phone, laptop and earbud makers offer whispered "private calls" and voice-anonymised "safe calls". Its own website describes assistive use as the next phase, one that will need medical validation. That gap, between the funded mission and the commercial route, is the main thing to watch in this grant.

How the whisper becomes a voice

Speech Technology magazine describes a whisper as voiceless speech, without a clear pitch, and Whispp's software as converting it into voiced speech. TNW reported in January 2024 that the company works audio to audio, avoiding the 2 to 3 seconds of delay typical of routes that first transcribe speech and then synthesise it, and that recordings of a user's earlier healthy voice let the output sound like that person. The December 2023 Speech Technology article listed the range of input it is meant to handle, from soft whispers to the rough speech left after a total laryngectomy, the removal of the voice box.

The stuttering case rests on a figure the company repeats: an average 85% reduction in stuttering when people who stutter severely whisper, cited by Speech Technology as research and by IO+ in February 2026. That number concerns whispering itself, not a clinical trial of Whispp; no peer-reviewed study of the product was found.

Because the conversion works on sound rather than words, Whispp calls it language independent, and the CORDIS objective says the same. The other claim that has moved over time is speed, and every figure below comes from the company or from articles quoting it.

From a Leiden pilot to a TIME list

In May 2021 Leiden University described Whispp as a start-up at the city's PLNT incubator, with linguistics students validating its audio data, a separate study with the Netherlands Cancer Institute and LUMC on voice disorders, and a three-month pilot with about 250 participants. In January 2024 TNW reported that the company was launching its first paid, subscription-based calling app and had closed a €750,000 seed round led by LUMO Labs, with Leiden University's holding Libertatis Ergo following on, a round also reported by The SaaS News. It was a CES 2024 Innovation Awards honoree in accessibility, and in October 2024 TIME listed it among its 200 Best Inventions of the year.

The record does not agree on when the company began. The 2021 Leiden article puts the start three years earlier, around 2018, and a November 2025 company release says 2018. TNW says 2019. A September 2026 company release and Capital Riesgo say 2020, which matches the registration date of 18 May 2020 shown by Liza.nl for the BV.

From an assistive app to device licensing

The commercial route changed in 2025 and 2026. In November 2025 Whispp appointed Andreas Spechtler, a former President of Dolby International and founder of a technology licensing business, to its board, and its CTO presented at a Qualcomm session on on-device AI at Web Summit, per the company's release. In July 2026 it announced a €5 million round led by LUMO Labs, which IO+ and Capital Riesgo both report as aimed at deployment with mobile and PC makers and semiconductor companies. On 22 September 2026 Spechtler became CEO and Castermans moved to Chief Marketing Officer, according to the company announcement.

The website now leads with WhisppWAVE kits for phones, Windows PCs and wearables, and two uses: speaking softly in an open office or public place while sounding normal to the listener, and hiding a caller's biometric voice behind an anonymised one. No device maker has been named as a licensee. LUMO Labs' Andy Lurling said the fund "initially invested in Whispp because of its mission to help people reclaim their voice", per Capital Riesgo, and Pulse 2.0 reports the company saying that mission remains central to its long-term vision.

The engine is the same technology the EIC funded, so this is a change of customer, not of product. It does mean that a project CORDIS describes in terms of voice disabilities is, in the company's current materials, sold first for privacy and discretion. The EIC side of the money also has a second part: alongside the grant, the EIC committed to an equity investment in a later phase, according to IO+ and Whispp. Neither gives an amount, and Whispp does not appear on the text of the EIC Fund beneficiaries page checked on 4 November 2026.

What can already be checked

The most concrete evidence is a six-week pilot with Qualcomm on Snapdragon X Elite laptops, published by Whispp and republished on Qualcomm's 3BL Media channel on 21 July 2026. It reports 30 participants in nine countries, a delay of about 48 to 50 milliseconds when the model ran on the chip's neural processing unit, and about 310 minutes of battery time against about 130 minutes on the main processor. In Whispp's own write-up, the pilot covered 13 voice conditions and 10 languages, 79% of participants rated it easy to use and 84% scored their likelihood of recommending it at 8 or more out of 10. Those survey figures appear only in the company's version, and no independent party is named as having run the measurements.

Little financial detail is public. Liza.nl, which reproduces register data, shows equity of €744,235 and debts of €1,658,657 in the latest year listed, 2025, up from equity of €296,545 in 2024, no turnover figure, and a staff figure it gives inconsistently as zero and one. These numbers could not be checked against the KvK filing itself, because the KvK search did not return the entry when this article was researched, so they are not charted. No revenue figure, licence fee or user count has been published by the company.

1. Real-time whisper-to-voice conversionExistsCalling app since 2024 per TNW and TIME; laptop pilot results published in 2026.
2. Fully on-device processingPilotedRun on a Snapdragon X Elite NPU in a 30-person pilot, per Whispp and Qualcomm.
3. Reach smartphones and laptops worldwideNo licensee namedSoftware kits listed on whispp.com; no shipping device announced.
4. Assistive use for voice disabilitiesNext phaseCompany says medical validation is still to come; no clinical study published.

Accessibility law and European voice AI

The European Accessibility Act, which has applied since 28 June 2025, sets accessibility requirements for smartphones, computers and electronic communications services sold in the EU. It does not require voice conversion, but it makes the accessibility of calling a product requirement for the phone and PC makers Whispp now targets. The CORDIS objective adds a second argument: the processing stays on the device, which the project presents as a matter of privacy and offline use as well as speed.

Whispp was not asked for comment for this article. The reporting is based on publicly available sources, listed below.