QA Engineer, On-Device Reconstruction Voice AI

About Whispp

Whispp is an award-winning deep-tech startup in Leiden. Our on-device Voice Reconstruction AI turns whispered, affected, and noisy speech into a person's clear, natural voice — in under 100 ms, on-device, with full speaker identity preserved. We license the technology as an SDK to smartphone and PC OEMs, chipset vendors, and hearing-protection manufacturers, and we have active proof-of-concept programs with tier-1 partners.

The role

You own the quality for the Whisper-to-Speech SDK — from the audio that comes out of it to the way it behaves on a customer's silicon. When a partner puts our technology in front of their users, your work is the reason they can trust it.

This role sits on two disciplines at once, and we need both. Embedded software testing tells you whether the SDK runs correctly on the target: latency budgets, memory, NPU behaviour, thermal limits, stability over hours of use. Audio and voice quality testing tells you whether the result is worth running at all: intelligibility, naturalness, preserved speaker identity, absence of artefacts. A build that measures clean but sounds wrong is a failure, and so is a convincing reconstruction that drops frames on the customer's chipset. We are looking for someone credible on both sides, who can catch what each discipline alone would miss.

This is our first dedicated QA hire. You will work at the intersection of our AI, engineering, and product teams: turning model behaviour into measurable quality criteria, turning quality data into product decisions, and turning what you find in the field into research priorities. You'll work directly with our AI researchers, integration and full-stack engineers, Technical Product Manager, and business development team, and you'll set the quality foundations that let us run multiple OEM programs at once.

What you'll do

Embedded software testing

Own the test strategy for the W2S SDK across devices, chipsets, and OS environments — Qualcomm Snapdragon, PC NPUs, emerging hearable platforms.

Define and maintain release criteria based on the PRD for each SDK version from v0.7 to v1.0, and hold the line on what ships.

Create and execute regression test plans for every SDK release: define scope and coverage, specify test cases and pass/fail criteria, run the plan across target hardware, and report results.

Build and run stress and soak tests covering real-time latency, memory, CPU/NPU load, thermal behaviour, and power draw.

Verify on-target behaviour that only shows up on real hardware: audio pipeline configuration, buffer underruns, scheduling jitter, backend and driver differences, recovery after interruptions.

Reproduce, isolate, and document defects precisely enough that engineering can act on them without a second round of questions.

Validate PoC and NRE deliverables against agreed KPIs before they reach the customer.

Audio and voice quality testing

Bring a trained ear and rigorous method to the perceptual quality of reconstructed speech: intelligibility, naturalness, speaker identity preservation, artefacts, and end-to-end latency.

Maintain evaluation pipelines built on established objective metrics (PESQ, ViSQOL, STOI, MCD, ASR-based WER) and complement them with structured listening tests (MOS, MUSHRA, ITU-T P.800 / P.808).

Build and curate test corpora covering whispered, affected, and pathological speech across speakers, languages, accents, microphone configurations, and acoustic environments.

Set up and maintain acoustic test rigs — reference microphones, artificial mouth / HATS, loopback latency measurement, controlled noise playback — so results are repeatable across devices, sites, and releases.

Judge quality the way the end user experiences it, not only the way the metric reports it, and make the case when the two disagree.

Cross-functional work

Feed real-world findings — noise conditions, microphone behaviour, latency, personalization edge cases — back to the AI team with the evidence needed to act on them.

Support the Technical Product Manager with the quality data behind roadmap, scoping, and go/no-go decisions.

Act as a technical quality contact for OEM engineering teams during validation, and produce test reports their engineers trust.

Keep demo hardware reliable and ready for customer meetings and trade shows.

Tooling and automation

Build automated test harnesses and hardware-in-the-loop setups that run against every SDK build.

Instrument device farms and CI so cross-platform regressions surface in hours, not at integration time.

Write repeatable QA playbooks, templates, and test plans that cut validation time for each new OEM program.

Maintain test and validation documentation alongside the SDK as it evolves.

What we're looking for

We consider embedded software testing and audio quality testing equally essential to this role. Strength in one and a genuine willingness to grow into the other will be considered; strength in neither will not.

Embedded and platform testing — required

5+ years in QA, test engineering, or validation for embedded, mobile, or real-time systems.

Hands-on Android testing: adb, instrumentation testing, logcat, systrace/Perfetto, device farms. Kotlin/Java is a plus.

Strong Python for test automation, tooling, and data analysis.

Experience validating on-device AI inference — NPU/DSP acceleration, TFLite, ONNX Runtime, or similar — including numerical parity across backends and quantization levels is a strong advantage.

Comfort working close to the hardware: profiling latency and memory, interpreting traces, and separating a platform problem from an SDK problem.

Audio and voice quality testing — equally required

A demonstrated affinity for audio and voice quality — you can hear the difference between a result that passes and one that actually sounds right to the user, and you can substantiate that judgment with measurement.

Real-time audio testing experience, and a solid grasp of Android audio pipelines (AudioRecord, AAudio, Oboe) and/or Windows audio frameworks (WASAPI, Windows Audio Session).

Practical audio measurement skills: end-to-end latency and jitter, glitch and dropout detection, signal-level analysis, objective quality metrics.

Ability to design listening evaluations that produce decisions rather than opinions.

Across both

Proven track record of writing and executing regression test plans, and of maintaining them as a product evolves.

Clear written communication and disciplined documentation of results.

Fluent written and spoken English; an affinity for languages, and ideally fluency in others.

Willing to travel to customer sites around 30% of the time.

Nice to have

Speech processing, voice AI, or audio enhancement background.

Experience designing and running formal listening tests (MUSHRA, MOS, P.808 crowdsourced evaluation).

Work with Qualcomm, MediaTek, or ARM platforms.

On-device model optimization, quantization, or latency profiling — and an understanding of what each costs in perceived quality.

Comfortable reading and debugging C/C++

Hearable, wearable, or hearing-protection validation.

GDPR and on-device biometric data handling.

CI/CD and test infrastructure ownership.

AOSP experience.

Your first year

A documented regression test plan exists for the SDK, and an automated regression suite runs against every build on at least two hardware targets.

An audio quality benchmark — objective metrics plus structured listening tests — is in place and tracked release over release.

Release criteria cover both on-device behaviour and perceived audio quality, are agreed across AI, engineering, and product, and are applied through v1.0.

A second OEM program is validated against documented, repeatable QA playbooks.

Send your CV and motivation to work@whispp.com.