# samples/

Two different things live here, and the distinction matters:

**`Cheetah_SpecDrum/`** — the SOURCE. WAV files, committed, the thing a fresh
clone needs to rebuild identical firmware.

**`smpNN.raw`** — build products. 8-bit signed mono 48kHz, generated by
`tools/importbank.py`, git-ignored. `tools/mksamples.py` bakes them into
`samples.h` at build time, which is also generated and also ignored.

## Rebuilding the bank

```sh
python tools/importbank.py samples/Cheetah_SpecDrum/<subset>
cmake --build build
```

`importbank.py` resamples to 48kHz, trims silence, matches loudness across
the whole bank, and dithers on the way down to 8 bits. See its docstring, and
`importwav.py` for why each of those steps is the way it is.

## A bank, not a per-voice list

Entry `NN` is not "voice NN's sound" — it is entry NN of a library. Which
voice plays which entry is a separate mapping, `kVoiceSample` in
`../samples_default.h`. That indirection is deliberate: the WebUI will
eventually let any entry be assigned to any of the twelve slots, and two
voices can share one recording without a second copy in flash.

So changing which sample a drum plays means editing `kVoiceSample`, not
re-importing audio.

## What is currently baked

Five of the Cheetah SpecDrum's `orig_*` kit:

| Bank | Sample | Voice slot |
|---|---|---|
| 0 | `orig_1_kick` | 0 kick |
| 1 | `orig_2_snare` | 2 snare |
| 2 | `orig_6_hihat_c` | 4 closed hat |
| 3 | `orig_7_hihat_o` | 6 open hat |
| 4 | `orig_8_clap` | 11 clap |

31KB, about 1.5% of flash. The other seven voices stay synthesised — a mix,
not a wholesale swap, which is the point of the card.

## Source

Cheetah SpecDrum sample ROMs (1986), from the ZX Spectrum drum machine of the
same name. Downloaded from **<https://samples.kb6.de/>**, which archives the
original ROM sets.

The `afro_`, `electro_`, `latin_` and `orig_` groups are its four factory
kits. Only five of `orig_` are baked in today; the other 27 files are here to
be swapped in by editing `kVoiceSample` — no re-import needed, since the bank
is indices rather than audio.