How detrimental is noise for your sleep? Andrej Karpathy raised this exact question in a tweet. The answer to Karpathy’s question is, well, empirical, with a rough experiment suggested by Paul Graham:
Consumer devices like Apple Watch measure many sleep metrics, including sleep stages, heart rate variability, and heart rate. The Apple Watch’s microphone also measures noise in fine-grained frequency, which can be filtered down to times when you’re asleep. Given that we’re a medical practice that makes extensive use of wearable data to improve patient care, and advance the science of heart health (where sleep and sleep apnea are major inputs), so we were curious to quantify the actual effect of noise on sleep.
TL;DR: across every metric we examined, bedroom noise shows a consistent, dose-dependent impact on sleep quality.
The effect of bedroom noise on six sleep metrics
Noise vs REM & deep sleep
Our panel shows a steady, almost linear erosion of REM minutes from the low-40 dB range through the mid-50s. Once noise climbs past roughly 58–60 dB (about the loudness of normal conversation), the line kinks sharply downward—REM time falls by an additional ~15 minutes in that single step. Deep-sleep minutes follow a similar pattern, holding fairly steady below 50 dB, slipping modestly through the low-50s, then dropping 6–7 minutes once the room crosses that same upper-50s threshold. Taken together, the curves suggest a threshold effect near 60 dB where restorative stages start to collapse rather than decline gradually.
Sleep Duration
Total sleep duration mirrors the stage-specific trends: nights under 55 dB hover around 6¼–6½ hours, but once noise breaches the upper-50s the average night shortens by nearly an hour—nearly all of it coming from REM and deep stages rather than lighter sleep.
Heart rate & HRV during sleep
Physiology responds in the opposite direction. Sleeping heart rate remains flat in the quietest bins, rises a beat or two in the low-50s, then jumps 4–5 bpm in the high-50s/low-60s. Sleeping HRV shows the mirror image: minor dips below 50 dB, followed by a pronounced 15–20 % drop once noise crosses that same ~60 dB mark. These shifts imply the body’s overnight stress load spikes at the same acoustic threshold that disrupts sleep architecture.
Empirical Health sleep score
Last, our own composite score—a weighted mix of duration, stages, and physiology—tracks the individual metrics closely. Scores stay in the high-70s below 55 dB but slide into the mid-60s once noise exceeds ~60 dB, confirming that the upper-50s/low-60s zone is the pivotal point where overall sleep quality degrades quickly.
Is there a threshold effect of noise on sleep?
The red dashed lines mark the single largest step-down for each series, and most of them cluster in a narrow band around 55–60 dB. Below that range, incremental quieting still buys modest gains, but above it the penalties accelerate: REM and deep minutes shrink by roughly a quarter, total sleep contract by about an hour, HR climbs by several beats per minute, and HRV flattens out.
Practically, that suggests a threshold effect: keeping bedroom sound levels beneath the low-60s dB (roughly the volume of normal conversation) is a pivotal target for preserving restorative sleep stages and the physiologic calm that goes with them.
