簡単なボイスチェンジャーのプログラム。

voice_changer.py

import sounddevice as sd
import numpy as np
import librosa
import blessed
import sys
import queue
import time
from scipy.signal import butter, sosfilt_zi, sosfilt

# --- 定数 ---
SAMPLE_RATE = 44100
CHANNELS = 1
BLOCK_SIZE = 2048
NOISE_GATE_THRESHOLD = 0.04
LOWCUT = 300.0
HIGCUT = 3400.0
FILTER_ORDER = 4
OUTPUT_GAIN = 0.5
FRAME_RATE = 20.0 # 描画のフレームレート

# --- ターミナルオブジェクト ---
term = blessed.Terminal()

# --- グローバル変数 ---
pitch_shift_semitones = 8.0
data_queue = queue.Queue()
sos_filter_coeffs = butter(FILTER_ORDER, [LOWCUT, HIGCUT], btype='band', fs=SAMPLE_RATE, output='sos')
zi_filter_state = sosfilt_zi(sos_filter_coeffs)

def get_color_from_amplitude(amplitude):
    """振幅に応じた色を返す"""
    normalized = min(1.0, amplitude * 4)
    if normalized < 0.3:
        return term.green
    elif normalized < 0.6:
        return term.yellow
    else:
        return term.red

def draw_waveform(data):
    """波形の描画に専念する関数"""
    global pitch_shift_semitones
    
    # ヘッダー表示
    header = f" VOICE CHANGER | Pitch Shift: {pitch_shift_semitones:.1f} semitones "
    print(term.move_y(0) + term.center(term.bold(header)))

    num_cols = term.width
    if data is None or data.size == 0:
        return
    
    step = len(data) // num_cols if len(data) > num_cols else 1
    if step == 0: step = 1
    sliced_data = data[::step]
    center_y = term.height // 2

    # 波形描画
    for x, value in enumerate(sliced_data):
        if x >= num_cols:
            break
        
        amplitude = int(value * (center_y - 2))
        color = get_color_from_amplitude(abs(value))

        for y in range(abs(amplitude)):
            pos_y = center_y - np.sign(amplitude) * y
            if 1 < pos_y < term.height:
                with term.location(x, int(pos_y)):
                    print(color('█'))

def callback(indata, outdata, frames, time, status):
    """音声処理とキューへのデータ投入に専念する関数"""
    global zi_filter_state, pitch_shift_semitones, data_queue

    if status:
        print(status, file=sys.stderr)

    try:
        mono_indata = indata.flatten()

        if np.abs(mono_indata).max() < NOISE_GATE_THRESHOLD:
            outdata.fill(0)
            if data_queue.qsize() < 2:
                data_queue.put(np.zeros_like(mono_indata))
            return

        filtered_data, zi_filter_state = sosfilt(sos_filter_coeffs, mono_indata, zi=zi_filter_state)

        shifted_data = librosa.effects.pitch_shift(
            y=filtered_data,
            sr=SAMPLE_RATE, 
            n_steps=pitch_shift_semitones
        )
        
        len_shifted = len(shifted_data)
        outdata_len = len(outdata.flatten())

        if len_shifted >= outdata_len:
            reshaped_data = shifted_data[:outdata_len].reshape(outdata.shape)
        else:
            padding = np.zeros(outdata_len - len_shifted)
            padded_data = np.concatenate((shifted_data, padding))
            reshaped_data = padded_data.reshape(outdata.shape)

        outdata[:] = reshaped_data * OUTPUT_GAIN
        
        if data_queue.qsize() < 2:
            data_queue.put(mono_indata)

    except Exception:
        outdata.fill(0)
        if data_queue.qsize() < 2:
            data_queue.put(np.zeros(BLOCK_SIZE, dtype=np.float32))

def main():
    """メインループで描画とキー入力を制御する関数"""
    global pitch_shift_semitones
    try:
        with sd.Stream(samplerate=SAMPLE_RATE, blocksize=BLOCK_SIZE, channels=CHANNELS, callback=callback):
            with term.fullscreen(), term.cbreak(), term.hidden_cursor():
                running = True
                last_data = np.zeros(BLOCK_SIZE, dtype=np.float32)
                
                while running:
                    # --- キューから最新のデータを取得 ---
                    try:
                        while not data_queue.empty():
                           last_data = data_queue.get_nowait()
                    except queue.Empty:
                        pass
                    
                    # --- 描画処理 ---
                    print(term.clear, end='')
                    draw_waveform(last_data)
                    
                    with term.location(0, term.height - 1):
                        controls = " [h] Pitch Up | [l] Pitch Down | [Enter] Exit "
                        print(term.center(term.bold_white_on_blue(controls)))
                    
                    sys.stdout.flush()

                    # --- キー入力とフレームレート制御 ---
                    key = term.inkey(timeout=1/FRAME_RATE)
                    if key:
                        if key == 'h':
                            pitch_shift_semitones = min(24.0, pitch_shift_semitones + 1.0)
                        elif key == 'l':
                            pitch_shift_semitones = max(-24.0, pitch_shift_semitones - 1.0)
                        elif key.name == "KEY_ENTER" or key == '\n':
                            running = False

    except Exception as e:
        # term.normalを呼び出してターミナルを通常状態に戻す
        print(term.normal)
        print(f"\nAn error occurred: {e}")
    finally:
        print("\nVoice changer stopped.")

if __name__ == "__main__":
    main()

requirements.txtを次のように修正します:

sounddevice
librosa
numpy
blessed
scipy
pyworld
cython