embedded-audio

no_std duty-modulated PWM audio for Cortex-M class MCUs: effect banks in external flash, tiered DSP, two-voice mixing, and crossfades.

Architecture

Flash (EAFX bank) → decode (Tier A/B/C) → 2 voices + ADSR → mix → ΣΔ → PWM duty

Call AudioEngine::tick() once per sample at AudioConfig::sample_rate_hz (default 16 kHz). Drive a fixed PWM carrier (default 32 kHz timer) by writing the returned duty compare value.

Quick start (firmware)

use embedded_audio::prelude::*;

static BANK: &[u8] = include_bytes!("../assets/ui.bank");

let bank = SoundBank::parse(BANK)?;
let mut engine = AudioEngine::new(AudioConfig::default_duty());
engine.set_bank(bank);
engine.play(1, AdsrSpec::click())?;

// Timer ISR @ 16 kHz:
let duty = engine.tick();
pwm.set_duty(duty);

Effect bank (EAFX v2)

Offset	Field
0..4	Magic EAFX
4	Version 2
5..6	Effect count (u16 LE)
7..8	Sample rate Hz (u16 LE)
9	Reserved
10+	Directory entries (16 bytes × N)
…	Payload bytes

Effect kinds

ID	Tier	Payload
Tone	A	empty — param0 = Hz, param1 = duration ms
Wavetable	A	256-byte table
Fm	A	empty — param0 = carrier Hz, param1 = mod ratio ×100
Pcm8	B	raw 8-bit mono
Adpcm	B	4-byte IMA header + nibbles
SigmaDeltaBits	C	MSB-first packed bits

Host baking

Multi-effect bank:

cargo run --features std --bin eaf-bake -- -o ui.bank --rate 16000 \
  --add 1:pcm8:click.raw \
  --add 2:adpcm:whoosh.raw \
  --add 3:tone:880:120 \
  --add 4:wavetable:440:table256.raw

Legacy single effect: eaf-bake --kind pcm8 --id 1 sound.raw -o bank.bin

Input for pcm8 / adpcm is unsigned 8-bit mono raw (center 128), band-limited in your DAW before export.
wavetable requires a 256-byte table file; param0 in the bank is playback Hz.

Preview (host)

cargo run --features std --bin eaf-preview -- --bank ui.bank --id 1 -o preview.wav

Uses tick_pcm() — the same mix/envelope path as firmware, before PWM mapping.

DMA buffer fill

let mut dma = [0u16; 128];
engine.fill_duty_buffer(&mut dma);
// DMA TX from `dma` at sample_rate_hz

Defaults

Constant	Value
DEFAULT_SAMPLE_RATE_HZ	16_000
DEFAULT_PWM_CARRIER_HZ	32_000
DEFAULT_PWM_PERIOD	1000 (set from your timer clock)
Output shaping	DutyMode::SigmaDelta

Real-time DSP & Analysis (dsp feature)

Enable the optional dsp feature to integrate zero-allocation digital signal processing algorithms powered by embedded-dsp (using libm for #![no_std] targets):

[dependencies]
embedded-audio = { version = "0.2.0", features = ["dsp"] }

use embedded_audio::prelude::*;
use embedded_audio::synth::Waveform;

let mut engine = AudioEngine::from_sample_rate(16000, 255, DutyMode::Linear);
engine.play_tone(440, 100, Waveform::Sine);

// 1. Equalize or filter engine audio with a Biquad filter (Lowpass, Highpass, Bandpass, Notch)
let mut filter = BiquadAudioFilter::lowpass(1000.0, 16000.0, 0.707);

let mut frame = [0.0f32; 256];
engine.fill_pcm_f32_buffer(&mut frame);
filter.process_buffer(&mut frame);

// 2. Measure audio signal statistics (RMS, Peak, Power, Mean, Variance)
let stats = AudioMeter::measure(&frame);
// stats.rms, stats.peak, stats.power, etc.

// 3. FFT Spectrum Analysis & Pitch / Dominant Frequency Detection
let (peak_freq_hz, peak_mag) = AudioSpectrumAnalyzer::find_peak_frequency(
    &frame,
    16000.0,
    WindowType::Hanning,
);

Features

Feature	Purpose
std	Host ADPCM encoder + eaf-bake binary
fm	Optional FM-buzzer backend (tick_fm, Markham profile) for bring-up only
dsp	Optional real-time DSP (BiquadAudioFilter, AudioSpectrumAnalyzer, AudioMeter, AudioLmsFilter) via embedded-dsp

RAM budget (typical)

• 2 voices + ADSR + ΣΔ state: < 512 B
• No heap; bank parsed from flash by reference

License

The contents of this repository are dual-licensed under the MIT OR Apache 2.0 License. That means you can choose either the MIT license or the Apache 2.0 license when you re-use this code. See LICENSE-MIT or LICENSE-APACHE for more information on each specific license.