How the sound is made

Every 2hives session is built in layers: a noise bed, a slowly moving filter on that bed, pairs of tones, and a voice. This page shows how each layer is made. The charts and numbers are measured on our own renders of one session, and you can listen to the layers one at a time.

Measured on: Lucid Dreaming, 30 minutes, rendered by the 2hives engine and delivered as FLAC and as AAC at 256 kbit/s. The design itself was set by measuring a reference curriculum of 36 tracks. Every chart and clip here comes from our own renders, not from those recordings. Method and reference measurements: research note.

26.8 dBvoice above the bed, median

9.9 sone sweep of the moving filter

21 / 21tone pairs found in the finished file

−31.6 LUFSintegrated loudness, delivered file

The signal chain

Noise, comb and tones are synthesised sample by sample. The voice is synthetic speech, placed in the centre. Only the voice is processed on its own; the bed is never filtered after it is made. A last, single gain sets the loudness of the whole mix.

The seven steps from a template and a script to a file. Steps 1 to 3 make the bed; step 4 is the voice; the mix and the mastering gain are applied once, at the end.

The noise bed

The bed is noise with more energy at low frequencies than at high ones. Noise that falls 3 dB per octave is called pink and noise that falls 6 dB per octave is called brown. Between 100 Hz and 3 kHz our default bed measures −6.4 dB per octave, so it is brown-leaning; with the colour setting on brighter it measures −4.9, and one course bed with a shaped spectrum measures −9.7.

Hover or tap the chart to read the levels.

Spectrum of three of our noise beds, in decibels relative to 1 kHz, against the 3 dB and 6 dB per octave reference lines. Measured on 60 seconds of rendered audio, ears averaged.

Data summary

Frequency, Hz	Default bed, dB	Brighter, dB	Course bed, dB
64	23.1	16.9	31.6
127	18.1	13.6	28.8
254	12.3	9.3	23.0
508	6.2	4.7	10.2
1016	−0.2	−0.1	0.0
2032	−7.3	−5.8	−8.9
4064	−16.2	−13.1	−14.6
8128	−27.2	−22.7	−31.6
16255	−41.2	−35.2	−34.0

In the phases with the moving filter, each ear gets its own noise. We measured the coherence between the left and right signal, from 0 (unrelated) to 1 (identical): 0.002 across 100 Hz to 10 kHz. In the opening phase the ears share most of their noise, and the same measurement gives 0.81.

Hover or tap the chart to read the coherence.

Left-right coherence of the noise by frequency: near zero in the phases with the moving filter, about 0.81 in the opening phase.

Data summary

Band	Movement phases	Opening phase
100 Hz – 1 kHz	0.002	0.81
1 – 4 kHz	0.002	0.81
4 – 10 kHz	0.002	0.81

The “ocean” movement

Add a copy of the noise that is delayed by a fraction of a millisecond and you get a comb filter: a row of notches across the spectrum. Change the delay and the notches slide. We sweep the delay smoothly, and each ear is swept a little out of step with the other, which gives the bed its sense of movement.

Measured on our render: the first notch travels between about 1.97 and 3.53 kHz once every 9.9 s, the right ear's sweep is ahead of the left ear's by 102 degrees, and the notch sits 16.6 dB below the surrounding level (the middle 80% of frames: 10.5 to 18.5 dB).

It is a movement in the spectrum, not a swell in volume. In the 2 to 4 kHz band each ear's level swings by ±2.1 dB over the cycle with the comb (±0.0 dB without it), and the left-right difference swings by ±3.2 dB.

Hover or tap the chart to read the notch frequencies.

The first notch of the comb in each ear over 50 seconds of a render. Both ears sweep the same range on the same 9.9 s cycle; the right ear is 102 degrees ahead.

Data summary

Measure	Value
Cycle length	9.89 s
First notch, lowest and highest (tracked)	1.97 – 3.62 kHz
First notch, lowest and highest (fitted sine)	1.97 – 3.53 kHz
Right ear ahead of left by	102°
Notch depth, median	16.6 dB

Position in the cycle

The filter's response in each ear as the sweep runs. This is an illustration drawn from the measured notch range, cycle, depth and lead above. It moves only while it is on screen, and it does not move at all if your system asks for reduced motion; the slider steps through the cycle by hand.

Tone pairs

A binaural pair is two sine tones, one in each ear, a few hertz apart. Stereo headphones carry each ear's channel separately, so each ear hears only its own tone; the gap between the two frequencies is the beat. Slow beats sit on low carriers and faster beats on higher ones.

A session stacks up to 6 pairs at a time. Each phase holds its own stack, and phases crossfade into one another over about a minute or two, so no pair starts or stops abruptly. The chart shows the Lucid Dreaming template: one row per pair, with carriers to the nearest 10 Hz and beats to the nearest 0.5 Hz, coloured by the band the beat falls in.

Hover, tap or tab to a row to read it.

Timeline of the 30-minute Lucid Dreaming template: each row is one tone pair, with slopes where it fades in and out.

Data summary

Carrier, Hz	Beat, Hz	Band	In, min	Out, min	Fade in / out, s
≈50	≈1	delta	7.9	30	≈60 / ≈10
≈100	≈4	theta	2.7	22.5	≈60 / ≈60
≈150	≈1.5	delta	7.9	30	≈60 / ≈10
≈160	≈10	alpha	0	3.4	≈10 / ≈45
≈200	≈1	delta	20.9	28.6	≈120 / ≈150
≈250	≈8	alpha	0	3.4	≈10 / ≈45
≈300	≈4	theta	2.6	16.9	≈45 / ≈45
≈400	≈4	theta	2.6	8.9	≈45 / ≈60
≈400	≈4	theta	16.1	22.9	≈45 / ≈120
≈510	≈10	alpha	16.1	22.9	≈45 / ≈120
≈600	≈7.5	theta	7.9	22.9	≈60 / ≈120

Phase	Band	From, min	To, min	Pairs found / sounding
Arrive	alpha	0	3	2 / 2
Loosen	theta	3	8.4	3 / 3
Lantern	theta	8.4	16.5	5 / 5
Intention	theta	16.5	21.9	6 / 6
Drift	delta	21.9	27.3	3 / 3
Hush	delta	27.3	30	2 / 2

We then analysed the finished file the way a spec sheet does. In the middle of each of the 6 phases we looked for tone pairs and compared them with what the engine was asked to play: 21 of 21 pairs were found, and the largest frequency error was 0.003 Hz.

The voice and the loudness

The voice is synthetic speech, placed in the centre: the same samples in both ears, so it stays put while the tones beat. It is the one layer processed on its own: a high-pass filter to remove rumble, a gentle de-esser, a level set for each section of the session, and a peak cap so that no line is louder than the cap.

Across the 17 lines of this session the voice sits 26.8 dB above the bed in the 300 Hz to 3.4 kHz speech band (range 23.9 to 31.4 dB). Measured over the whole spectrum the gap is 15.2 dB, because the bed keeps most of its energy below 300 Hz. Peaks are held at −3.0 dBFS. Lines are a median 49 s apart and speech fills 3.9% of the session, so most of the time you hear sound alone.

Hover or tap a line to read it.

Each stem is one spoken line, placed by start time; its height is how far the voice sits above the bed in the speech band.

Data summary

Start, min	Length, s	Above the bed, dB	Phase
0.1	4.4	30.2	arrive
0.4	4.5	23.9	arrive
0.5	4.9	27.1	arrive
0.6	1.2	27.4	arrive
0.7	4.3	27.5	arrive
7.1	3.8	24.1	loosen
8.6	1.6	27.4	lantern
8.7	4.3	27.2	lantern
10.1	5.0	26.1	lantern
10.3	5.1	26.8	lantern
16.7	4.5	26.3	intention
17.0	5.8	25.5	intention
18.1	1.9	25.3	intention
18.8	6.1	25.0	intention
20.1	3.5	26.5	intention
22.2	5.8	29.9	drift
23.3	4.2	31.4	drift

Hover or tap the chart to read the levels.

Level of the speech and of the bed in one-third-octave bands. The speech sits well above the bed across the band where it carries meaning.

Data summary

Band centre, Hz	Voice, dBFS	Bed, dBFS
100	−57.8	−43.5
126	−53.0	−54.7
159	−37.4	−50.5
200	−27.4	−56.5
252	−30.4	−57.5
317	−38.0	−51.1
400	−31.4	−59.8
504	−34.2	−61.1
635	−33.9	−59.3
800	−34.0	−64.1
1008	−34.3	−66.2
1270	−37.7	−68.9
1600	−39.2	−72.5
2016	−40.1	−77.4
2540	−41.5	−80.6
3200	−44.4	−79.0
4032	−45.7	−78.5
5080	−46.3	−80.5

The finished file measures −31.6 LUFS integrated, a loudness range of 16.2 LU and a true peak of −2.0 dBTP. Mastering applies one static gain, aiming at −28 LUFS but never past a true peak of −2 dBTP. This session needed 4.6 dB to reach the target; the peak ceiling allowed 1.0 dB, so the ceiling set the result.

Hover or tap the chart to read the loudness.

Short-term loudness (a 3-second window) of the delivered FLAC through the session, with the integrated loudness marked.

Does it survive compression?

A download comes as lossless FLAC and as AAC at 256 kbit/s, which is 4.6 times smaller for this session (265 MB against 57 MB). We took a 32-second excerpt, encoded it both ways and ran the same analysis on the decoded files.

All 5 tone pairs were found in both files. The largest frequency error was 0.003 Hz after FLAC and 0.003 Hz after AAC; our research note counts a beat as reproduced at 0.02 Hz or less. The comb survives too: median notch depth 16.6 dB in FLAC and 16.3 dB in AAC (a change of −0.3 dB), the notch tracks differ by a median of 0.0 Hz, and the right ear's lead measures 102 and 102 degrees.

Hover or tap a pair to read its errors.

Frequency error of each detected tone pair after FLAC and after AAC 256 kbit/s, against a line at 0.02 Hz, the limit our research note uses for a reproduced beat.

Data summary

Carrier, Hz	Beat, Hz	FLAC error, Hz	AAC error, Hz
≈300	≈4	0.000	0.000
≈150	≈1.5	0.000	0.000
≈50	≈1	0.000	0.000
≈600	≈7.5	0.000	0.000
≈100	≈4	0.003	0.003

Hover or tap the chart to read the notch frequencies.

The comb's first notch in the left ear after FLAC and after AAC 256 kbit/s: the two tracks lie on top of each other.

Listen layer by layer

Four clips of 15 seconds, rendered from the same seed and the same moment of the session (about minute 10). Each adds one layer to the one before. They are encoded as AAC at 256 kbit/s, like a download.

Start with your volume low. The first three layers are quiet, as in the real file, and the voice layer is about 12 dB louder.

1 / 4 · Noise only · 445 KB

2 / 4 · Noise + swept comb · 446 KB

3 / 4 · Noise + comb + tone pairs · 447 KB

4 / 4 · Noise + comb + tone pairs + voice · 474 KB

Plain audio elements: no processing in the page, nothing plays until you press play, and starting one clip stops the others. Switching layer keeps your place in the clip.

What is measured, and what is not known

Measured on our renders

The spectrum, coherence and movement of the noise bed, the frequencies and fades of the tone pairs, the level of the voice against the bed, the loudness and peak of the delivered file, and how all of these come through FLAC and AAC.
One template (Lucid Dreaming, 30 minutes), one voice and one seed. Other templates differ in their stacks and phases.

Not measured, and not known

We did not measure any effect on listeners. Nothing here shows what a session does for you.
Binaural beats are real: two slightly different tones, one per ear, create a perceived beat in the brainstem (Oster, 1973). Whether they “entrain” brainwaves is unclear: a 2023 systematic review found 5 of 14 EEG studies consistent with entrainment, 8 contradicting it and 1 mixed (Ingendoh et al., 2023).
Some studies report small effects on cognitive measures and others find none, and context matters too: guidance and expectation all shape the experience. The evidence is mixed, and we publish both sides.
We do not know whether the moving filter changes anything beyond how the bed sounds, or how other voices, speakers or playback chains compare with these measurements.

Read the full evidence summary and the papers on our research page, including the 2023 systematic review (PDF). 2hives sessions are for relaxation, meditation and self-exploration, not a substitute for professional care. See the disclaimer.

Spec sheets

Every course has a spec sheet that lists what is in its audio: the phases, carriers, beats and noise, checked against the files themselves. Each render ships with its own copy.

Last measured 2026-10-12. The page, its charts and its clips are rebuilt from the engine's own renders by a script, so the numbers in the text always match the figures.