Olympus WS-853 Digital Voice Recorder

A digital voice recorder converts air-pressure variations at the microphone into an electrical signal, samples it, and stores it as a digital audio file. In paranormal use these recorders are run to capture Electronic Voice Phenomena (EVP) — short, speech-like sounds (typically one to three syllables, often whispered and low in volume) found on a recording when no speaker was consciously heard at the time. Because EVP are usually quiet relative to background noise, they are normally surfaced afterwards by raising the gain or filtering the file. Related instruments in this category work on the same audio signal in different ways: an amplified EVP microphone boosts very low-level sound before it reaches the recorder; an EVP spike / sound meter triggers a light or reading when sound level crosses a threshold; and an ultrasound monitoring device (Para4ce UMD) shifts frequencies above human hearing down into the audible band so they can be heard or logged. PRN treats a recording as a record of sound, not of speech. Deciding that a captured fragment is a word is an interpretive act and is where most error enters.

To record EVP at all, there must be noise in the audio circuit — and raising gain to find faint sounds also raises that noise floor, so artefacts and genuine quiet sounds are amplified together. Recorders are susceptible to radio-frequency interference (RFI). Inadequately shielded devices can pick up broadcast transmissions or stray electromagnetic signals, which is why serious EVP work is sometimes done in RF- and sound-screened rooms. Auditory pareidolia is the dominant risk. The brain readily imposes familiar words onto ambiguous sound, especially when primed with an expectation ("say your name"). A fragment that "clearly" says something to a listener who knows the prompt may carry no such content. Built-in mics capture handling and body noise — fabric rustle, footsteps, breathing, stomach noise, distant traffic and HVAC — all of which can resemble whispers once amplified. Spike/sound meters do not identify a source. They register that a sound was loud enough, not what made it.

Stray radio / broadcast pickup through poor shielding (RFI). Investigator-generated sound — whispers, breathing, clothing, equipment handling, footsteps. Building and environmental noise — pipes, HVAC, wind, settling timber, insects, traffic. Amplification artefacts — compression, clipping and noise-floor hiss introduced when gain is raised to recover a faint sound. Pareidolia at review — assigning words to noise, especially when the reviewer expects a particular answer. Best practice: log every audible sound at the time ("tagging"), use the same file for everyone's review, avoid telling reviewers what a clip is "supposed" to say, prefer recorders with good RF shielding, and keep a noise baseline of the room. A captured sound is evidence of a sound; it is not, by itself, evidence of a voice.