XIAO ESP32S3でESP Audio Effectsを流用してみる-思考とメイカーの狭間

https://www.notyet-maker.com/2026/01/xiao-esp32s3-7.html

の続き。

これまで、ChatGPTに質問しながら、XIAO ESP32S3でエレキギターっぽい音を出そうとしていたが、一向に進まなかった。

Geminiに質問したところ、ESP Audio Effectsというものがあることを知る。

またもやうまく進まないのではないか、と心配だけど、ChatGPTに質問しながら、

CircuitPythonでESP Audio Effectsの設計思想を流用する、ということについてトライしてみる。

具体的には、CircuitPythonで、

発音：synthioで基音だけを作る

Fader：amplitudeやゲインを時間で動かす

Mixer：audiomixerでノイズ層と基音層を混ぜる

Sonic：audiodelaysで微小ディレイや揺れを足す

という対応に置き換える、という。

しばらく、ChatGPTとやり取りしていたが、解決しなさそうなので、

Geminiの方が優れているのか質問したら、非を認めてしまった。

今回は、Gemini3を利用することにする。

ノイジーだけど、Smoke on the Waterっぽい音を作ることができた。

https://youtube.com/shorts/HTuSfzI2nQ4?si=Ccn9mSG48dEIBDX5

///

import math

import time

from array import array

import audiobusio

import board

import synthio

# --- Hardware / runtime ---

SAMPLE_RATE = 18000

I2S_BCLK = board.D6

I2S_LRCLK = board.D3

I2S_DATA = board.D7

# --- Sequencer (Smoke on the Water riff, 16th-note timing) ---

BEAT_SEC = 0.134

G2 = 98.00

BB2 = 116.54

C3 = 130.81

DB3 = 138.59

NOTE_SEQUENCE_ABSOLUTE = [

(0, G2, 2),

(3, BB2, 2),

(5, C3, 7),

(16, G2, 2),

(19, BB2, 2),

(21, DB3, 2),

(23, C3, 6),

(32, G2, 2),

(35, BB2, 2),

(37, C3, 2),

(40, BB2, 2),

(42, G2, 10),

]

LOOP_STEPS = 48

RELEASE_TAIL_SEC = 0.05

NOTE_VELOCITY = 0.8

# --- Sound layer (synthio) ---

AMP_MIN = 0.06

AMP_MAX = 0.08

ATTACK = 0.003

DECAY = 0.40

SUSTAIN = 0.85

RELEASE = 0.12

UPDATE_STEP = 0.01

ATTACK_WAVE_MS = 12

ATTACK_DECAY_MS = 40

ATTACK_PITCH_MS = 18

PITCH_CENTS = 2.8

LP_FC_NORMAL = 4800.0

LP_FC_DAMP = 1500.0

SAT_DRIVE = 0.6

POST_GAIN = 0.78

CAB_FC = 4200.0

CAB_CUT = 0.55

CAB_MID_DRIVE = 0.30

WAVE_LEN = 128

MAX_I16 = 32767

SUSTAIN_AMP_BOOST = 2.20

AMP_SOFT_CLIP = 0.10

SUSTAIN_BRIGHT_MIX = 0.36

SUSTAIN_DARK_MIX = 0.12

SUSTAIN_WOBBLE_MAX_CENTS = 0.06

SUSTAIN_WOBBLE_STEP = 0.02

SUSTAIN_WOBBLE_DELAY_MS = 60

SUSTAIN_WOBBLE_PROB = 0.10

VEL_BINS = (0.35, 0.65, 1.0)

HARMONIC_PROFILE_GAIN = 0.28

HARMONIC_PROFILE = (

(2, -0.22),

(3, 0.30),

(4, -0.18),

(5, 0.20),

(6, -0.12),

)

DRIVE_BASE = 0.5

ODD3_BASE = 0.65

ODD5_BASE = 0.18

PHASE_WARP_BASE = 0.00

# --- Noise burst layer (synthio note) ---

NOISE_WAVE_LEN = 256

NOISE_BURST_MIN_MS = 4

NOISE_BURST_MAX_MS = 8

NOISE_LEVEL_MAX = 0.08

rng = 1

def rand_unit():

global rng

rng = (rng * 1103515245 + 12345) & 0x7FFFFFFF

return rng / 0x7FFFFFFF

def clamp01(value):

if value < 0.0:

return 0.0

if value > 1.0:

return 1.0

return value

def cents_to_ratio(cents):

return 2 ** (cents / 1200.0)

def alpha_from_fc(fc):

dt = 1.0 / SAMPLE_RATE

rc = 1.0 / (2.0 * math.pi * fc)

return dt / (rc + dt)

def soft_sat(x, drive):

y = x + drive * (x * x * x)

if y > 1.0:

y = 1.0

elif y < -1.0:

y = -1.0

return y

def clamp_i16(value):

if value > MAX_I16:

return MAX_I16

if value < -MAX_I16:

return -MAX_I16

return value

def soft_clip_amp(value):

if value <= AMP_SOFT_CLIP:

return value

excess = value - AMP_SOFT_CLIP

return AMP_SOFT_CLIP + (excess / (1.0 + (excess * 8.0)))

def mix_waves(wave_a, wave_b, mix):

w = array("h", [0] * WAVE_LEN)

a = 1.0 - mix

b = mix

for i in range(WAVE_LEN):

w[i] = clamp_i16(int((wave_a[i] * a) + (wave_b[i] * b)))

return w

def vel_key(value):

best = VEL_BINS[0]

best_dist = abs(value - best)

for level in VEL_BINS[1:]:

dist = abs(value - level)

if dist < best_dist:

best = level

best_dist = dist

return best

WAVE_CACHE = {}

def get_waves(velocity):

key = vel_key(velocity)

waves = WAVE_CACHE.get(key)

if waves is not None:

return waves

drive = DRIVE_BASE + (0.5 * key)

wave_normal = make_wave(drive, ODD3_BASE, ODD5_BASE, PHASE_WARP_BASE, LP_FC_NORMAL)

wave_attack = make_wave(

drive * 1.3,

ODD3_BASE * 1.25,

ODD5_BASE * 1.15,

PHASE_WARP_BASE,

LP_FC_NORMAL,

)

wave_damp = make_wave(drive, ODD3_BASE, ODD5_BASE, PHASE_WARP_BASE, LP_FC_DAMP)

wave_mid = mix_waves(wave_normal, wave_attack, SUSTAIN_BRIGHT_MIX)

wave_sustain = mix_waves(wave_normal, wave_attack, SUSTAIN_DARK_MIX)

waves = (wave_normal, wave_attack, wave_mid, wave_sustain, wave_damp)

WAVE_CACHE[key] = waves

return waves

def make_wave(drive, odd3, odd5, phase_warp, lp_fc):

w = array("h", [0] * WAVE_LEN)

lp_alpha = alpha_from_fc(lp_fc)

lp_y = 0.0

cab_alpha = alpha_from_fc(CAB_FC)

cab_y = 0.0

for i in range(WAVE_LEN):

phase = 2.0 * math.pi * i / WAVE_LEN

phase = phase + phase_warp * math.sin(phase)

x = math.sin(phase)

for harm, weight in HARMONIC_PROFILE:

x += HARMONIC_PROFILE_GAIN * weight * math.sin(harm * phase)

x = x + (odd3 * drive) * (x * x * x) + (odd5 * drive) * (x * x * x * x * x)

x = soft_sat(x, SAT_DRIVE)

lp_y = lp_y + lp_alpha * (x - lp_y)

cab_y = cab_y + cab_alpha * (lp_y - cab_y)

shaped = (lp_y * (1.0 - CAB_CUT)) + (cab_y * CAB_CUT)

shaped = soft_sat(shaped, CAB_MID_DRIVE)

z = shaped * POST_GAIN

if z > 1.0:

z = 1.0

elif z < -1.0:

z = -1.0

w[i] = int(z * MAX_I16)

return w

def build_noise_wave():

data = array("h", [0] * NOISE_WAVE_LEN)

for i in range(NOISE_WAVE_LEN):

n = (rand_unit() * 2.0) - 1.0

data[i] = clamp_i16(int(n * MAX_I16))

return data

i2s = audiobusio.I2SOut(I2S_BCLK, I2S_LRCLK, I2S_DATA)

synth = synthio.Synthesizer(sample_rate=SAMPLE_RATE)

i2s.play(synth)

env = synthio.Envelope(

attack_time=ATTACK,

decay_time=DECAY,

sustain_level=SUSTAIN,

release_time=RELEASE,

)

noise_wave = build_noise_wave()

noise_env = synthio.Envelope(

attack_time=0.001,

decay_time=0.006,

sustain_level=0.0,

release_time=0.008,

)

noise_state = None

def trigger_noise_burst(now):

global noise_state

burst_ms = NOISE_BURST_MIN_MS + rand_unit() * (NOISE_BURST_MAX_MS - NOISE_BURST_MIN_MS)

noise_note = synthio.Note(

frequency=440.0,

envelope=noise_env,

amplitude=NOISE_LEVEL_MAX,

waveform=noise_wave,

)

synth.press(noise_note)

noise_state = {

"start": now,

"end": now + (burst_ms / 1000.0),

"note": noise_note,

}

def update_noise(now):

global noise_state

if noise_state is None:

return

if now >= noise_state["end"]:

synth.release(noise_state["note"])

noise_state = None

return

def fader_gain(base_amp, t_ms):

if t_ms < ATTACK_WAVE_MS:

return min(base_amp * 1.15, AMP_MAX)

if t_ms < ATTACK_DECAY_MS:

blend = (t_ms - ATTACK_WAVE_MS) / (ATTACK_DECAY_MS - ATTACK_WAVE_MS)

return base_amp * (1.15 - (0.15 * blend))

return soft_clip_amp(base_amp * SUSTAIN_AMP_BOOST)

def play_note(freq, velocity, now):

base_amp = AMP_MIN + (AMP_MAX - AMP_MIN) * velocity

wave_normal, wave_attack, wave_mid, wave_sustain, wave_damp = get_waves(velocity)

freqs = (freq, freq * 1.3348)

states = []

for voice_freq in freqs:

note = synthio.Note(

frequency=voice_freq, envelope=env, amplitude=base_amp, waveform=wave_attack

)

synth.press(note)

states.append(

{

"note": note,

"start": now,

"amp": base_amp,

"freq": voice_freq,

"wave_attack": wave_attack,

"wave_mid": wave_mid,

"wave_sustain": wave_sustain,

"wave_damp": wave_damp,

"wobble_value": (rand_unit() - 0.5) * SUSTAIN_WOBBLE_MAX_CENTS * 0.5,

}

)

return states

def update_note(state, now):

note = state["note"]

t = now - state["start"]

t_ms = t * 1000.0

if t < (ATTACK_WAVE_MS / 1000.0):

note.waveform = state["wave_attack"]

elif t_ms < ATTACK_DECAY_MS:

note.waveform = state["wave_mid"]

elif t_ms < 800.0:

note.waveform = state["wave_sustain"]

else:

note.waveform = state["wave_damp"]

note.amplitude = fader_gain(state["amp"], t_ms)

if t < (ATTACK_PITCH_MS / 1000.0):

decay = math.exp(-t / 0.01)

note.frequency = state["freq"] * cents_to_ratio(PITCH_CENTS * decay)

else:

if t >= (SUSTAIN_WOBBLE_DELAY_MS / 1000.0):

if rand_unit() < SUSTAIN_WOBBLE_PROB:

wobble_value = state["wobble_value"] + (

(rand_unit() - 0.5) * 2.0 * SUSTAIN_WOBBLE_STEP

)

if wobble_value > SUSTAIN_WOBBLE_MAX_CENTS:

wobble_value = SUSTAIN_WOBBLE_MAX_CENTS

elif wobble_value < -SUSTAIN_WOBBLE_MAX_CENTS:

wobble_value = -SUSTAIN_WOBBLE_MAX_CENTS

state["wobble_value"] = wobble_value

note.frequency = state["freq"] * cents_to_ratio(state["wobble_value"])

else:

note.frequency = state["freq"]

current_notes = None

event_index = 0

loop_offset_steps = 0

start_time = time.monotonic() + 0.2

next_event_time = start_time + (NOTE_SEQUENCE_ABSOLUTE[0][0] * BEAT_SEC)

note_off_time = 0.0

current_hold = 0.0

while True:

now = time.monotonic()

if current_notes is None and now >= next_event_time:

selected = None

while now >= next_event_time:

start_step, freq, hold_units = NOTE_SEQUENCE_ABSOLUTE[event_index]

hold_sec = max(0.0, (BEAT_SEC * hold_units) - RELEASE_TAIL_SEC)

selected = (freq, hold_sec)

event_index += 1

if event_index >= len(NOTE_SEQUENCE_ABSOLUTE):

event_index = 0

loop_offset_steps += LOOP_STEPS

next_event_time = start_time + (

(loop_offset_steps + NOTE_SEQUENCE_ABSOLUTE[event_index][0]) * BEAT_SEC

)

if selected is not None:

freq, hold_sec = selected

current_notes = play_note(freq, NOTE_VELOCITY, now)

current_hold = hold_sec

note_off_time = now + hold_sec

trigger_noise_burst(now)

elif current_notes is not None and now >= note_off_time:

for state in current_notes:

synth.release(state["note"])

current_notes = None

if current_notes is not None:

for state in current_notes:

update_note(state, now)

update_noise(now)

time.sleep(UPDATE_STEP)

///

コードの説明文は、ChatGPTの方が良かった。

---

全体としては「Smoke on the Water を16分刻みの“絶対時刻”で鳴らしつつ、ギターっぽい発音を synthio の波形切替＋揺れ＋ノイズで作る」コードです。

音を出す土台

audiobusio.I2SOut(D6,D3,D7) で I2S 出力し、synthio.Synthesizer(sample_rate=18000) を i2s.play(synth) で常時駆動しています。

リフ（リズムと音程）の指定方法 NOTE_SEQUENCE_ABSOLUTE は「開始ステップ・周波数・保持ステップ数」を持つ配列で、BEAT_SEC=0.134 を1ステップとして扱います。 start_time + (loop_offset_steps + start_step)*BEAT_SEC により絶対時刻で再生するため、処理遅延があってもテンポが崩れにくい設計です。

“ギターっぽさ”の作り方（音色）

make_wave() で倍音・歪み・サチュレーション・簡易キャビネット特性を含む周期波形を生成し、get_waves() でアタック／中間／サステイン／減衰後用の波形を用意します。再生中は update_note() が経過時間に応じて note.waveform を切り替え、明るさが徐々に落ちる挙動を作っています。

発音の“それっぽさ”（挙動）

fader_gain() でアタック直後を強調しつつ伸びを作り、ATTACK_PITCH_MS の間だけピッチをずらしてピック直後の張力変化を表現します。サステイン中は確率的な微小ピッチ揺れで、機械的な音を避けています。

2音同時（このリフの太さ）

freqs = (freq, freq * 1.3348) により基音＋完全4度付近の2音を同時発音し、Smoke on the Water 特有のダブルストップ感を出しています。

ノイズバースト（ピックのアタック）

ノート開始時に短いノイズを重ね、ピックの瞬間的な成分を補っています。

メインループ（スケジューラ）

while now >= next_event_time で遅れを吸収しつつイベントを選択し、note_off_time で離鍵、再生中は update_note() を毎ループ適用します。

---

XIAO ESP32S3でESP Audio Effectsを流用してみる

0 件のコメント:

コメントを投稿

このブログを検索

ブログアーカイブ

ラベル

不正行為を報告