SimpleSampler

A low-latency sample pad that lives in your terminal. Load any WAV, map it to a key or MIDI note, and play — no DAW required.

Built for live performance where every millisecond counts.

Features

• 4x4 pad grid in the terminal via Textual — keyboard, mouse, or MIDI trigger
• ~6ms audio latency — 256-sample buffer with lock-free voice mixing, no mutexes in the audio path
• MIDI input — bind any note, CC, or program change to any pad with a simple string like note:36:ch9
• Hot-swap MIDI devices from within the app — press m, pick a device, keep playing
• Instant playback — samples preloaded into memory as float32 NumPy arrays (up to 10 MB cache)
• Supports 8/16/24-bit WAV, mono or stereo, any sample rate (auto-resampled to 44.1kHz)

Requirements

• Linux (this is all we care about)
• Python >= 3.13
• PortAudio — the system library that makes audio go fast

  # Debian / Ubuntu
  sudo apt install libportaudio2
  # Arch
  sudo pacman -S portaudio
  

• uv — fast Python package manager

Installation

git clone https://github.com/yourusername/SimpleSampler.git
cd SimpleSampler
uv sync

That's it. No virtualenv juggling, no pip drama.

Usage

uv run simplesampler my_bank.json

The app launches in your terminal with a 4x4 pad grid. Smash keys, click pads, or send MIDI — audio fires instantly.

Controls

Key	Action
1 2 3 4	Pads 0-3 (top row)
q w e r	Pads 4-7
a s d f	Pads 8-11
z x c v	Pads 12-15
m	Open MIDI device selector
Mouse click	Trigger any pad

Keybinds are customizable per-pad in the bank JSON (see below). The defaults above apply when no keybind is set.

Bank Configuration

A bank is a JSON file that defines your pad layout. Here's a minimal example:

{
"name": "My Kit",
"midi_device": "MPD218",
"pads": [
{
"id": 0,
"name": "Kick",
"sample_path": "/path/to/kick.wav",
"color": "red",
"keybind": "1",
"midibind": "note:36:ch9"
},
{
"id": 1,
"name": "Snare",
"sample_path": "/path/to/snare.wav",
"color": "green",
"keybind": "2",
"midibind": "note:38:ch9"
}
]
}

Bank fields

Field	Type	Required	Description
name	string	yes	Display name for the bank
midi_device	string	no	MIDI input device — name substring (e.g. "MPD218") or index (e.g. "0"). Omit to skip MIDI on startup.
pads	array	yes	List of pad objects (up to 16)

Pad fields

Field	Type	Required	Description
id	int	yes	Pad position, 0-15 (4x4 grid, left-to-right, top-to-bottom)
name	string	yes	Display label on the pad
sample_path	string	yes	Absolute path to a WAV file
color	string	no	Pad color — any Textual color (default: "blue")
keybind	string	no	Keyboard key to trigger this pad (default: positional 1234qwerasdfzxcv)
midibind	string	no	MIDI binding string (see below)

Midibind format

note:<number>:ch<channel> — MIDI note (e.g. note:36:ch9)
cc:<number>:ch<channel> — Control change (e.g. cc:1:ch0)
pc:<number>:ch<channel> — Program change (e.g. pc:5:ch0)

Don't know your MIDI note numbers? That's what midi_learn.py is for.

MIDI Learn

midi_learn.py is a standalone CLI tool that listens on a MIDI device and prints midibind strings to stdout as you hit pads/keys/knobs. Copy the output straight into your bank JSON.

# List available MIDI devices
uv run python midi_learn.py -l
# Listen on a specific device (by name)
uv run python midi_learn.py -d "MPD218"# Listen on a specific device (by index)
uv run python midi_learn.py -d 0
# Auto-select first available device
uv run python midi_learn.py

Hit a pad on your controller, get a string:

note:36:ch9
note:38:ch9
note:42:ch9
cc:1:ch0

Paste those into your bank JSON's midibind fields. Ctrl+C when you're done.

How It's Fast

A few deliberate choices that add up:

• sounddevice + PortAudio — 256-sample blocks at 44.1kHz = ~5.8ms buffer. Reports actual achieved latency on startup.
• Lock-free audio — voices are queued via collections.deque (atomic append in CPython). The audio callback drains its own local list. No mutex ever touches the hot path.
• Direct key dispatch — on_key fires audio through a pre-built dict lookup, completely bypassing Textual's async widget message queue. You hear the sound before the pad even flashes.
• MIDI daemon thread — polls at 2ms intervals via iter_pending(), triggers audio directly from the thread. No round-trip through the UI event loop.
• Preloaded samples — WAV files are decoded, resampled, and converted to float32 stereo NumPy arrays at startup. Zero file I/O during playback.

License

This project is licensed under the GNU General Public License v3.0 — see the LICENSE file for details.