Mountain Brook is the steadiest water recording we have: its loudness varies by only 0.31 dB and it has no sudden peaks at all. It was recorded in 2026 on a small stream in the Tian Shan mountains near Almaty. This page shows the numbers, the one flaw we found, and exactly what we did to the audio before it became a long track.
The recording
| Place | Mountain brook, Tian Shan, near Almaty |
|---|---|
| Year | 2026 |
| Recorder | Zoom H5, WAV 48 kHz / 24-bit |
| Part used | 0:20–19:55 of the take |
| Analysed | 85 s excerpt |
| Video | Tinnitus Relief Sound Masking · Mountain Brook |
The spot we almost skipped
In photos, this place looked like a splashy little cascade. Splashes mean impacts, and impacts are what we try to avoid in a sleep track, so we nearly rejected it. When we measured the take, it turned out to be the most even sound in our library. The rocks break the flow into fine foam, and foam sounds like a smooth hiss with no individual hits.
That changed how we work. We never judge a spot by how it looks any more. We record, then measure. More on how we choose places is in our field notes from the Tian Shan.
Measurements
| Measure | Result | What it means |
|---|---|---|
| Loudness variation | 0.31 dB | Almost no change in level from moment to moment |
| Spread (P95–P5) | 1.04 dB | Nearly all moments sit within about 1 dB of each other |
| Sudden peaks >6 dB | 0 frames | No splashes, knocks or clicks above the flow |
| Drift across the excerpt | 0.2 dB RMS | The level does not creep up or down |
| Tone centre | 1115–1136 Hz | The pitch of the hiss holds still |
| Tonal peaks | none | No hum or whistle hiding in the sound |
| Energy below 40 Hz | 0.059% | No wind rumble or traffic drone |
| Clipping | none | The recorder never overloaded |
| Loop seam | 0.00 dB | No level jump where the loop repeats |
| Left / right balance | L +3.5 dB | Fixed See below |
Loudness variation is the standard deviation of the level in 400 ms windows. A sudden peak is a window more than 6 dB above the median level. Tone centre is the spectral centroid. Left/right correlation was 0.58.
For comparison, a dawn birdsong recording measured the same way had 3.73 dB of loudness variation and 42 loud events per minute (full birdsong analysis). Why steadiness matters for masking is explained in our tinnitus masking guide.
The flaw: one side was louder
The only problem the measurements found was balance. The left channel was 3.5 dB louder than the right. On speakers you would hardly notice. In headphones, over a whole night, the sound would sit noticeably to one side. We corrected it with a pan filter before looping, so both ears get the same level.
How we prepared the track
- Trim the spoken slate: we kept 0:20 to 19:55 of the take.
- Balance the channels to remove the 3.5 dB lean to the left.
- Soften the top end with a high shelf of −5 dB at 6 kHz.
- No compressor. The brook is already steady, and compression would only lift background noise.
- Normalise the peaks to −3 dBFS.
- Loop to the full track length, with a checked seam of 0.00 dB.
No sound was generated or added. What you hear is the brook, trimmed, balanced and slightly softened.
Listen
More recordings, grouped by sound, are in the library.