{author: hhr, date: 191216, function: comment}

Poz:

I also enjoyed the "lurking" feeling of reading other parallel conversations

:) The same here. This is of course the difference with cooperate platforms that monitor what you read - here we never see what you read where and when… it only becomes apparent in the writing.

**HHR: notes 1**

---

meta: true

author: HHR

artwork: ThroughSegments

project: AlgorithmicSegments

---

{author: hhr, date: 191217, kind: note, keywords: [sweep, impulse response, space, interaction]}

Probing the space. What if I sent out tiny sweep burst to measure impulse responses, deliberately picking up all the smearing from interactions in the space?

Following the experiments with statistical filtering, I added arithmetic mean and spectral flatness features to the filter calculation, which can be used as additional selectors, passing through only certain chunks. I implemented a Histogram UGen for FScape (this will go into the next release version of Mellite), to understand the dynamic of these features. I find it interesting that histograms are basically **counting** mechanisms, so I can see a relation to Ji's ideas about counting steps.

{author: hhr, date: 200111, function: contextual, keywords: [statistical, filter, counting, histogram]}

I experimented with feeding the fragments through the minimum phase segmentation, but this wasn't good yet, as I need better range information for the segmentation. Instead, I now feed it through a strong limiter to flatten the loudness differences between the fragments, followed by a "paul stretch" time stretch; currently with fixed factor 8, but this should be determined by the signal itself. Also it should probably allow overlapping segments.

What I like very much about this is that there is a simple control now over the **density of events**. Instead of basing the thresholds on one fixed histogram measurement, probably that measurement should also be done at intervals by the system.

{author: hhr, date: 200111, kind: caption, keywords: [filter, minimum phase, segmentation, stretch, density]}

Histogram of arithmetic mean of spectral frames of the filter function obtained from a microphone recording. There are 256 bins which where taken from the log-arithmetic mean in the range of -10 to 0.0. The filter function is (A - B) / B, where A = long term median window of spectral magnitudes, B = short term median window of spectral magnitudes.

{author: hhr, date: 200111, kind: caption, keywords: [statistical, filter, counting, histogram]}

Histogram of geometric mean of the same signal. There are 256 bins which where taken from the log-geometric mean in the range of -50 to 0.0.

{author: hhr, date: 200111, kind: caption, keywords: [statistical, filter, counting, histogram]}

Histogram of spectral flatness (geometric mean divided by arithmetic mean) of the same signal. There are 256 bins which where taken from the log-geometric mean in the range of -50 to 0.0.

{author: hhr, date: 200111, kind: caption, keywords: [statistical, filter, counting, histogram]}

Corresponding Mellite workspace

{kind: repository}

{hhr, 200118}

Regarding foreground/background, this analysis from my thesis p.301 still seems very good (TO-DO: put the text here as text):

{author: hhr, date: 200111, kind: caption}