| Filename | Latest commit message | Latest commit date |
|---|---|---|
| .. | ||
| README.md | ||
Bird Syrinx Golf
This sidequest uses open-licensed bird recordings as a reality check for the graph-native syrinx path. The goal is not to make species-accurate bird impersonations yet. The goal is to force the paired-source syrinx graph to meet clean calls, emit listenable WAVs, and expose which controls currently matter.
Source Set
All references are Wikimedia Commons files, with Commons file pages used as the license/provenance anchor and Commons MP3 transcodes used for robust local decoding. The Korean National Institute of Biological Resources clips are KOGL Type 1 rather than Creative Commons; keep that attribution distinction visible.
| id | species | call | author | license | source |
|---|---|---|---|---|---|
common-iora-xc125847 |
Common Iora (Aegithina tiphia) |
song | Sudipto Roy | CC BY-SA 3.0 | https://commons.wikimedia.org/wiki/File:Aegithina_tiphia_-_Common_Iora_XC125847.ogg |
common-blackbird-1059970 |
Common Blackbird (Turdus merula) |
song | Diana Tudor | CC BY 4.0 | https://commons.wikimedia.org/wiki/File:Common_Blackbird_song_(Turdus_merula).ogg |
red-footed-falcon |
Red-footed Falcon (Falco vespertinus) |
typical calls | Bubulcus | CC BY 3.0 | https://commons.wikimedia.org/wiki/File:Falco_vespertinus.ogg |
warbling-white-eye-ko |
Warbling White-eye (Zosterops japonicus) |
call | National Institute of Biological Resources | KOGL Type 1 | https://commons.wikimedia.org/wiki/File:%EB%8F%99%EB%B0%95%EC%83%88.ogg |
eurasian-wren-ko |
Eurasian Wren (Troglodytes troglodytes) |
song | National Institute of Biological Resources | KOGL Type 1 | https://commons.wikimedia.org/wiki/File:%EA%B5%B4%EB%9A%9D%EC%83%88.ogg |
Tool
tools/BirdSyrinxGolf downloads the sources, extracts the loudest 1.2 second window, renders a small grid of graph-native Aqua syrinx candidates, and writes:
reference-clip.wavcandidate-syrinx.wavcandidate-syrinx.aquasource.jsonreport.txt- run-level
summary.md
The generated audio lives under artifacts/bird-syrinx-golf/<timestamp>/, which is intentionally ignored by git.
Clean Reference Run
Run: artifacts/bird-syrinx-golf/20260528T143401133
| bird | score | logMelCosine | best coarse control |
|---|---|---|---|
| Common Iora | 0.2423 | 0.2356 | freq=1050, high beak opening |
| Common Blackbird | 0.2784 | 0.3223 | freq=1050, high beak opening |
| Red-footed Falcon | 0.2727 | 0.3004 | freq=3000, narrow labial opening |
| Warbling White-eye | 0.3102 | 0.3692 | freq=3000, narrow labial opening |
| Eurasian Wren | 0.2577 | 0.2081 | freq=1050, high beak opening |
Interpretation
This proves the current graph source path can render paired-labium syrinx candidates against clean real bird audio. It does not prove the model is good. The search grid is deliberately coarse and mostly static; it cannot yet match call envelopes, syllable timing, pitch sweeps, side-to-side alternation, beak articulation, or turbulence bursts with enough authority.
The useful finding is that the syrinx controls now form an inspectable fitting surface. The next coherent step is not a larger brute-force grid. It is time-varying control curves over the same graph: pressure envelopes, labial tension sweeps, independent left/right gating, beak opening, and possibly tract morphology changes when the reference call has obvious articulatory motion.