EVT Guide: Acoustic Isolation & PCB Design Layout
An engineering validation method for separating acoustic, mechanical, electrical, and software failures inside a compact two-microphone voice recorder.

EVT should turn a compact recorder from a collection of working subsystems into one measurable system. The correct method is to isolate acoustic, mechanical, electrical, radio, power, firmware, and transcription variables—then reproduce each failure before changing the design.
What must EVT prove for a compact voice recorder?
C1 combines two MEMS microphones, a moving recording indicator, a 1250 mAh battery, storage, Bluetooth 5.4, USB-C, and local computer-side ASR inside a 60 × 64 × 20 mm enclosure. EVT must establish whether those parts can operate simultaneously without violating the acoustic path, power integrity, radio link, thermal envelope, or data lifecycle.
Begin with a block-level failure tree. “Bad transcript” is not a root cause. It can originate in low acoustic signal-to-noise ratio, clipping, clock drift, vibration, electrical coupling, packet loss, decoding, voice activity detection, the ASR model, or incorrect language selection.
How do you separate acoustic noise from electrical noise?
Use controlled substitutions. Record the same calibrated or repeatable speech source in at least four states: board outside the enclosure, board in the enclosure, turntable disabled, and turntable active. Repeat on battery and while charging. If the artifact follows the enclosure or motor state, investigate acoustic transmission and vibration. If it follows charging or digital activity, investigate power and electromagnetic coupling.
Microsoft’s audio hardware guidance identifies both external/internal acoustic noise and electrical noise as received-noise sources. It recommends treating microphone placement and integration as primary speech-recognition factors. Practical EVT measurements include:
- noise floor with the sound source off;
- frequency response with a fixed source and geometry;
- clipping margin at near speech levels;
- channel gain and phase matching between the two microphones;
- spectral peaks correlated with motor speed, DC/DC switching, USB activity, or radio bursts;
- transcript critical-term recall on the same recorded corpus.
Save raw captures and configuration metadata. A screenshot of a favorable waveform is not enough to reproduce a result.
How should PCB layout protect the microphone path?
The exact layout depends on whether the MEMS microphones provide analog or digital output, but the review logic is stable. Keep sensitive microphone routes short, provide an intentional return path, separate them from switching nodes and motor current, decouple at the device pins, and avoid routing across gaps in the reference plane. Do not place a high-current return so that it shares impedance with the microphone reference.
For digital microphones, clock edge integrity and return current still matter. “Digital” does not mean immune to power noise, radiated coupling, or timing errors. For analog microphones, gain-stage noise, bias stability, and ADC reference quality become additional variables.
Use a current probe, near-field probe, oscilloscope, or temporary firmware modes to correlate artifacts with subsystems. An EVT board should provide test points for rails, reset, clocks, microphone data, motor drive, and ground. Removing every test point to save space can make the next failure much more expensive to diagnose.
How do you test structure-borne vibration from the turntable?
Test with the same speech source while the mechanism is off, starting, at steady state, and stopping. A slow mechanism can still produce tonal energy from gear mesh, eccentricity, bearing contact, or pulse-width-modulated drive. Compare microphone spectra and time-domain events with a reference accelerometer or contact sensor on the housing if available.
Change one mechanical variable at a time: motor mount, isolation element, gear preload, part balance, clearance, or microphone gasket. If multiple changes are made together, a quieter result cannot identify which change mattered or whether a new reliability risk was introduced.
What should the EVT test matrix contain?
| Axis | Required states | Why it matters |
|---|---|---|
| Acoustic scene | Quiet speech, keyboard noise, multiple speakers, near and approximately 3 m | Exercises signal level and interference |
| Mechanism | Off, start, steady, stop | Separates transient and periodic vibration |
| Power | Battery, USB power, low battery | Finds rail and charger coupling |
| Transfer | Bluetooth, USB-C, interruption, reconnect | Checks data integrity and recovery |
| Thermal | Cold start, steady operation, charging | Finds drift and fit changes |
| Software | VAD on/off test mode, decoder versions, fixed ASR model | Prevents software changes from masking hardware causes |
How do you close an EVT issue?
An issue closes only when the team has a reproducible setup, before/after data, an identified design change, the change in the controlled source files, and a regression test. A bench wire or foam patch is diagnostic evidence, not a released fix.
Track the measurement uncertainty and sample count. One quiet unit cannot establish manufacturing capability. EVT establishes mechanisms and margins; DVT validates the complete behavior, and PVT checks repeatability at production intent.
Sources and further reading
- Microsoft audio-device design guidance — received noise, microphone arrays, phase matching, bandwidth, and speech-recognition targets.
- Microsoft AEC and microphone-array sample — official Core Audio example separating capture, stream management, and voice processing.
- IETF RFC 6716: Opus — standards-track definition of the speech and audio codec used in the documented OneMira data design.
- OneMira C1 specifications — approved public hardware and local-processing facts.