python 音頻可視化

　　代碼整理好放在 github 上了： https://github.com/darkchii/visualize

　　bilibili 演示視頻：https://www.bilibili.com/video/av77372866

　　2020-07-18 11:50:05 更新一個線條顏色漸變的方法：

import matplotlib.pyplot as plt
import numpy as np
import pyaudio
from _tkinter import TclError
from pydub import AudioSegment
from matplotlib.animation import FuncAnimation
from matplotlib.collections import LineCollection


p = pyaudio.PyAudio()
sound = AudioSegment.from_file(file='F:/Music/中原めいこ - Cloudyな午后.mp3')
left = sound.split_to_mono()[0]
fs = left.frame_rate
size = len(left.get_array_of_samples())
channels = left.channels
stream = p.open(
    format=p.get_format_from_width(2,),
    channels=1,
    rate=50000,  # 調整播放速率
    # input=True,
    output=True,
)
stream.start_stream()
fig = plt.figure()
ax = fig.gca(
    # projection='polar'
)
norm2 = plt.Normalize(-1., 1.)
lc = LineCollection([], cmap='gist_ncar', norm=norm2)
ax.set_ylim(-1.5, 1.5)
ax.set_axis_off()
window = int(0.02*fs)
freq = np.linspace(20, 20000, window // 2)
time = np.linspace(0, 20, window)
ax.add_collection(lc)


def update(frames):
    if stream.is_active():
        slice = left.get_sample_slice(frames, frames + window)
        stream.write(slice.raw_data)
        y = np.array(slice.get_array_of_samples()) / 30000
        points = np.array([time, y]).T.reshape(-1, 1, 2)
        segments = np.concatenate([points[:-1], points[1:]], axis=1)
        lc.set_segments(segments)
        lc.set_array(y)

    return lc,


ani = FuncAnimation(fig, update, frames=range(0, size, window), interval=0, blit=True)
plt.show()

　　運行截圖：

　　2020-03-23 13:23:01 還是給一個 pydub（需要自己配置好ffmpeg）正確使用姿勢，因為 mp3 格式太常見：

import numpy as np
import pyaudio
from pydub import AudioSegment, effects
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation


p = pyaudio.PyAudio()
sound = AudioSegment.from_file(file='../xxx.mp3')
left = sound.split_to_mono()[0]
fs = left.frame_rate
size = len(left.get_array_of_samples())
channels = left.channels
stream = p.open(
    format=p.get_format_from_width(left.sample_width,),
    channels=channels,
    rate=fs,
    # input=True,
    output=True,
)

stream.start_stream()
fig = plt.figure()
ax1, ax2 = fig.subplots(2, 1)
ax1.set_ylim(0, 0.5)
ax2.set_ylim(-1.5, 1.5)
ax1.set_axis_off()
ax2.set_axis_off()
window = int(0.02*fs) # 20ms
f = np.linspace(20, 20*1000, window // 2)
t = np.linspace(0, 20, window)
lf1, = ax1.plot(f, np.zeros(window // 2), lw=1)
lf2, = ax2.plot(t, np.zeros(window), lw=1)


def update(frames):
    if stream.is_active():
        slice = left.get_sample_slice(frames, frames + window)
        data = slice.raw_data
        stream.write(data)
        y = np.array(slice.get_array_of_samples()) / 30000 # 歸一化
        yft = np.abs(np.fft.fft(y)) / (window // 2)

        lf1.set_ydata(yft[:window // 2])
        lf2.set_ydata(y)

    return lf1, lf2,


ani = FuncAnimation(fig, update, frames=range(0, size, window), interval=0, blit=True)
plt.show()

　　截圖：

　　2020-02-25 14:50:27 Animation：

　　注：pyaudio open 調節參數 format 有驚喜（取值范圍{1, 2, 4, 8, 16,...}）

　　極坐標版：

import matplotlib.pyplot as plt
from scipy.signal import detrend
# from scipy.fftpack import fft
import numpy as np
import pyaudio
from _tkinter import TclError
import struct
import wave
# import librosa
from pydub import AudioSegment
from matplotlib.animation import FuncAnimation


chunk = 1024
p = pyaudio.PyAudio()
# sound = AudioSegment.from_file(file='../Music/xxx.mp3')
# rdata = sound.get_array_of_samples()
wf = wave.open('../Music/xxx.wav')
stream = p.open(
    format=8,
    channels=wf.getnchannels(),
    rate=wf.getframerate(),
    # input=True,
    output=True,
    # frames_per_buffer=chunk
)

fig = plt.figure()
ax = fig.gca(projection='polar')
# ax.set_ylim(0, 1)
ax.set_axis_off()
lf, = ax.plot(np.linspace(0, 2 * np.pi, chunk), np.zeros(chunk), lw=1)


def init():
    stream.start_stream()
    return lf,


def update(frame):
    if stream.is_active():
        data = wf.readframes(chunk)
        stream.write(data)
        data_int = struct.unpack(str(chunk * 4) + 'B', data)
        y_detrend = detrend(data_int)
        yft = np.abs(np.fft.fft(y_detrend))
        y_vals = yft[:chunk] / (chunk * chunk * 4)
        ind = np.where(y_vals > (np.max(y_vals) + np.min(y_vals)) / 2)
        y_vals[ind[0]] *= 2
        lf.set_ydata(y_vals)
    return lf,


ani = FuncAnimation(fig, update, frames=None,
                    init_func=init, interval=0, blit=True)
plt.show()

　　截圖：

　　stem 版：

import matplotlib.pyplot as plt
from scipy.signal import detrend
# from scipy.fftpack import fft
import numpy as np
import pyaudio
from _tkinter import TclError
import struct
import wave
# import librosa
from pydub import AudioSegment
from matplotlib.animation import FuncAnimation


chunk = 1024
p = pyaudio.PyAudio()
# sound = AudioSegment.from_file(file='../Music/xxx.mp3')
# rdata = sound.get_array_of_samples()
wf = wave.open('../Music/xxx.wav')
stream = p.open(
    format=8,
    channels=wf.getnchannels(),
    rate=wf.getframerate(),
    # input=True,
    output=True,
    # frames_per_buffer=chunk
)

fig = plt.figure()
ax = fig.gca()
ax.set_ylim(0, 1)
ax.set_axis_off()
lf = ax.stem(np.linspace(20, 20000, chunk), np.zeros(chunk), basefmt=':', use_line_collection=True)
lf.markerline.set_color([0.8, 0.2, 0, 0.5])


def init():
    stream.start_stream()
    return lf


def update(frame):
    if stream.is_active():
        data = wf.readframes(chunk)
        stream.write(data)
        data_int = struct.unpack(str(chunk * 4) + 'B', data)
        y_detrend = detrend(data_int)
        yft = np.abs(np.fft.fft(y_detrend))
        y_vals = yft[:chunk] / (chunk * chunk)
        ind = np.where(y_vals > (np.max(y_vals) + np.min(y_vals)) / 2)
        y_vals[ind[0]] *= 4
        lf.markerline.set_ydata(y_vals)
    return lf


ani = FuncAnimation(fig, update, frames=None,
                    init_func=init, interval=0, blit=True)
plt.show()

　　semilogx 版：

import matplotlib.pyplot as plt
from scipy.signal import detrend
# from scipy.fftpack import fft
import numpy as np
import pyaudio
from _tkinter import TclError
import struct
import wave
# import librosa
from pydub import AudioSegment
from matplotlib.animation import FuncAnimation


chunk = 1024
p = pyaudio.PyAudio()
# sound = AudioSegment.from_file(file='../Music/xxx.mp3')
# rdata = sound.get_array_of_samples()
wf = wave.open('../Music/xxx.wav')
stream = p.open(
    format=8,
    channels=wf.getnchannels(),
    rate=wf.getframerate(),
    # input=True,
    output=True,
    # frames_per_buffer=chunk
)

fig = plt.figure()
ax = fig.gca()
ax.set_ylim(0, 1)
ax.set_axis_off()
lf, = ax.semilogx(np.linspace(20, 20000, chunk), np.zeros(chunk), lw=1, color='lightblue')


def init():
    stream.start_stream()
    return lf,


def update(frame):
    if stream.is_active():
        data = wf.readframes(chunk)
        stream.write(data)
        data_int = struct.unpack(str(chunk * 4) + 'B', data)
        y_detrend = detrend(data_int)
        yft = np.abs(np.fft.fft(y_detrend))
        y_vals = yft[:chunk] / (chunk * chunk)
        ind = np.where(y_vals > (np.max(y_vals) + np.min(y_vals)) / 2)
        y_vals[ind[0]] *= 4
        lf.set_ydata(y_vals)
    return lf,


ani = FuncAnimation(fig, update, frames=None,
                    init_func=init, interval=0, blit=True)
plt.show()

　　2020-02-25 12:18:49 更新一下：

　　這次更新是給一個很勉強的通過音頻數據來播放音樂並展示 fft 效果的代碼，實際上僅僅只是播放音樂很簡單，wav 波形文件只需要 wave 即可，想要 mp3 或者其他支持較廣類型的音頻文件格式，使用 pydub 包更好.。（更新：使用 pydub 不需要解包，只需要通過 get_array_of_samples() 就可以獲取數據，raw_data 是字節數組，用於加載到輸出流）

　　但還是給個代碼：

import matplotlib.pyplot as plt
from vispy.plot import Fig
from matplotlib.colors import LightSource
from matplotlib import cm
from scipy.signal import detrend
from scipy.fftpack import fftn
import numpy as np
import pyaudio
from _tkinter import TclError
import struct
import wave
# import librosa
import array
from pydub import AudioSegment
from pydub.utils import get_array_type


chunk = 1024

p = pyaudio.PyAudio()

# sound = AudioSegment.from_file(file='../Music/1563833285950.mp3')
# left = sound.split_to_mono()[1]
# bit_depth = left.sample_width * 8
# array_type = get_array_type(bit_depth)
# numeric_array = array.array(array_type, left.raw_data)

wf = wave.open('../Music/1563833285950.wav')
stream = p.open(
    format=p.get_format_from_width(wf.getsampwidth()),
    channels=wf.getnchannels(),
    rate=wf.getframerate(),
    # input=True,
    output=True,
    # frames_per_buffer=chunk
)
stream.start_stream()
freq = np.linspace(20, 20000, chunk)
yf = np.zeros(chunk)
fig = plt.figure()
ax = fig.gca()
lf, = ax.semilogx(freq, yf, lw=1, color='lightblue')
ax.set_ylim(0, 1)
ax.set_axis_off()
plt.show(block=False)

start = 0
while stream.is_active():
    data = wf.readframes(chunk)
    if len(data) < chunk:
        break
    stream.write(data)

    data_int = struct.unpack(str(chunk * 4) + 'B', data)

    y_detrend = detrend(data_int)
    yft = np.abs(fftn(y_detrend))
    y_vals = yft[:chunk] / (128 * chunk)

    lf.set_ydata(y_vals)

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　給個 pydub 將數據合成到包的源碼截圖：

　　2019-12-31 15:45:34 再來一個：
　　極坐標版：

import matplotlib.pyplot as plt
from matplotlib.colors import LightSource
from matplotlib import cm
from scipy.signal import detrend
import numpy as np
import pyaudio
from _tkinter import TclError
import struct


channels = 1
rate = 48000
chunk = 2048
p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    frames_per_buffer=chunk
)
stream.start_stream()
theta = np.linspace(0.0, rate, chunk)
radii = np.zeros(chunk)
fig = plt.figure()
ax = fig.gca(projection='polar')
lf = ax.stem(theta, radii, basefmt=':', use_line_collection=True)
lf.markerline.set_color([0.8, 0.2, 0, 0.5])
ax.set_rorigin(0)
ax.set_axis_off()
plt.show(block=False)
while stream.is_active():
    data = stream.read(chunk)
    data_int = struct.unpack(str(chunk * 2) + 'B', data)
    y_detrend = detrend(data_int)
    yft = np.abs(np.fft.fft(y_detrend))
    y_vals = yft[:chunk] / (128 * chunk)
    lf.markerline.set_ydata(y_vals)

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　笛卡爾直角坐標版：

import matplotlib.pyplot as plt
from matplotlib.colors import LightSource
from matplotlib import cm
from scipy.signal import detrend
import numpy as np
import pyaudio
from _tkinter import TclError
import struct


channels = 1
rate = 48000
chunk = 2048
p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    frames_per_buffer=chunk
)
stream.start_stream()
theta = np.linspace(0.0, rate, chunk)
radii = np.zeros(chunk)
fig = plt.figure()
ax = fig.gca()
lf = ax.stem(theta, radii, basefmt=':', use_line_collection=True)
lf.markerline.set_color([0.8, 0.2, 0, 0.5])
ax.set_ylim(0, 1)
ax.set_axis_off()
plt.show(block=False)
while stream.is_active():
    data = stream.read(chunk)
    data_int = struct.unpack(str(chunk * 2) + 'B', data)
    y_detrend = detrend(data_int)
    yft = np.abs(np.fft.fft(y_detrend))
    y_vals = yft[:chunk] / (128 * chunk)
    lf.markerline.set_ydata(y_vals)

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　運行截圖：

　　2019-12-30 14:05:57 再來更新一個版本（可惜我的電腦跑起來非常卡）：

from scipy.signal import detrend
import numpy as np
import pyaudio
from _tkinter import TclError
import struct


channels = 1
rate = 48000
chunk = 2048

p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    frames_per_buffer=chunk
)

stream.start_stream()
theta = np.linspace(0.0, 2. * np.pi, chunk, endpoint=False)
radii = np.zeros(chunk)
fig = plt.figure()
ax = fig.gca(projection='polar')
lf = ax.bar(x=theta, height=radii, width=0.02, bottom=-2.0, alpha=0.5)
ax.set_axis_off()
plt.show(block=False)

while stream.is_active():
    data = stream.read(chunk)
    data_int = struct.unpack(str(chunk * 2) + 'B', data)
    y_detrend = detrend(data_int)
    yft = np.abs(np.fft.fft(y_detrend))
    y_vals = yft[:chunk] / (128 * chunk)

    for rect, y_val, color in zip(lf.patches, y_vals, plt.cm.Spectral(y_vals * 8)):
        rect.set_height(y_val)
        rect.set_facecolor(color)

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　2019-11-28 20:58:07 更新，圍繞在圓環上的2d 版（一開始sb了，直接加在 y 軸上 = =，幸好馬上意識到需要做向量旋轉）：

from _tkinter import TclError
import pyaudio
import numpy as np
import matplotlib.pyplot as plt
import struct
from scipy.signal import savgol_filter, detrend, lfilter


channels = 1
rate = 48000
chunk = 1024 * 2

p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    # output=True,
    frames_per_buffer=chunk
)

stream.start_stream()

R = 2
t = np.linspace(0, 2*np.pi, chunk * 2)
xf = R*np.cos(t)
yf = R*np.sin(t)
fig, ax = plt.subplots(figsize=(7, 7))
lf, = ax.plot(xf, yf, lw=1)
ax.set_xlim(-3, 3)
ax.set_ylim(-3, 3)
ax.set_axis_off()
plt.show(block=False)

fwhm = 20

while stream.is_active():
    data = stream.read(chunk)
    data_int = struct.unpack(str(chunk * 2) + 'B', data)
    y_detrend = detrend(data_int)
    # z_smooth = savgol_filter(data_int, window_length=len(data_int) - 1, mode='valid')
    # box = np.ones(fwhm) / fwhm
    # z_smooth = np.convolve(y_detrend, box, mode='valid')
    yft = np.abs(np.fft.fft(y_detrend))
    y_vals = yft / (256 * chunk)
    ind = np.where(y_vals > np.mean(y_vals))
    y_vals[ind[0]] *= 4
    lf.set_xdata(xf + y_vals * np.cos(t))
    lf.set_ydata(yf + y_vals * np.sin(t))

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　運行截圖：

　　2019-11-28 16:47:13 更新，3D 版：

from _tkinter import TclError
import pyaudio
import numpy as np
import matplotlib.pyplot as plt
import struct
from scipy.signal import savgol_filter, detrend, lfilter


channels = 1
rate = 48000
chunk = 1024 * 2

p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    # output=True,
    frames_per_buffer=chunk
)

stream.start_stream()
R = 20
t = np.linspace(0, 2*np.pi, chunk * 2)
xf = R*np.cos(t)
yf = R*np.sin(t)
# fig, ax = plt.subplots(figsize=(14, 5))
fig = plt.figure()
ax = fig.gca(projection='3d')
lf, = ax.plot(xf, yf, np.zeros(chunk * 2), lw=2)
# ax.set_xlim(20, rate / 2)
# ax.set_ylim(20, rate / 2)
ax.set_zlim(-0.2, 1.2)
ax.set_axis_off()
plt.show(block=False)

fwhm = 20

while stream.is_active():
    data = stream.read(chunk)
    data_int = struct.unpack(str(chunk * 2) + 'B', data)
    z_detrend = detrend(data_int)
    # z_smooth = savgol_filter(data_int, window_length=len(data_int) - 1, mode='valid')
    # box = np.ones(fwhm) / fwhm
    # z_smooth = np.convolve(z_detrend, box, mode='valid')
    zf = np.abs(np.fft.fft(z_detrend))
    z_vals = zf / (256 * chunk)
    ind = np.where(z_vals > np.mean(z_vals))
    z_vals[ind[0]] *= 4
    lf.set_xdata(xf)
    lf.set_ydata(yf)
    lf.set_3d_properties(z_vals)

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　運行截圖：

　　23:27:12 更新，為了更加突出重要的頻率，我又加了一些代碼，就是把振幅大於平均值的那些振幅加倍，主要代碼如下：

y_vals = yf[:chunk] / (256 * chunk)
ind = np.where(y_vals > np.mean(y_vals))
y_vals[ind[0]] *= 4
lf.set_ydata(y_vals)

　　22:17:15 更新，因為 python 3.8 和 scipy 暫時不兼容，所以用不了scipy 對信號做平滑處理，不過網上找了一段代碼，效率不錯，這是鏈接，或者直接把下面代碼加到合適的地方。

width = 20
box = np.ones(width) / width
y_smooth = np.convolve(data_int, box, mode='same')
yf = np.fft.fft(y_smooth)
...

　　21:58:43 更新，測試了一下對原始信號做平滑處理，發現效率十分低，完全沒法看效果，不過目前用的是自己實現的高斯濾波算法。。。

以下為原文

　　這里的簡單原理就是獲取輸入輸出設備中的數據（注意驅動什么的沒有問題，能用麥克風），然后 matplotlib 繪制出來，想要看到音樂的節奏振動就再 fft 一下。至於如何不斷更新波形，matplotlib 有一個 animation 方法可以用（見下面第二種方法），但實際上我用了之后發現顯示效果不如第一種（可能是姿勢不對）。之前用 matlab 做的，也很不錯。

　　第一種方法（波形顯示更流暢，代碼參考這個視頻）：

from _tkinter import TclError
import pyaudio
import numpy as np
import matplotlib.pyplot as plt
import struct


channels = 1
rate = 44100
chunk = 1024 * 2

p = pyaudio.PyAudio()

stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    output=True,
    frames_per_buffer=chunk
)

stream.start_stream()

xf = np.linspace(0, rate, chunk)
fig, ax = plt.subplots()
lf, = ax.semilogx(xf, np.zeros(chunk), '-', lw=1)
ax.set_xlim(20, rate / 2)
ax.set_ylim(0, 1)
plt.show(block=False)

while stream.is_active():
    data = stream.read(chunk)
    data_int = struct.unpack(str(chunk * 2) + 'B', data)
    yf = np.fft.fft(data_int)
    lf.set_ydata(np.abs(yf[:chunk]) / (128 * chunk))

    try:
        ax.figure.canvas.draw()
        ax.figure.canvas.flush_events()
    except TclError:
        stream.stop_stream()
        stream.close()
        break

　　第二種方法：

import pyaudio
import numpy as np
# from scipy.fftpack import fft
import matplotlib.pyplot as plt
import struct
from matplotlib.animation import FuncAnimation


channels = 1
rate = 44100
chunk = 1024 * 2
p = pyaudio.PyAudio()
stream = p.open(
    format=pyaudio.paInt16,
    channels=channels,
    rate=rate,
    input=True,
    output=True,
)

stream.start_stream()
x = np.arange(0, 2*chunk, 2)
xf = np.linspace(0, rate, chunk)
fig, (ax1, ax2) = plt.subplots(2)
l, = ax1.plot(x, np.zeros(chunk), '-', lw=1)
lf, = ax2.semilogx(xf, np.zeros(chunk), '-', lw=1)
ax1.set_xlim(0, 2*chunk)
ax1.set_ylim(0, 255)
ax2.set_xlim(20, rate / 2)
ax2.set_ylim(0, 1)
plt.setp(ax1, xticks=[0, chunk, 2 * chunk], yticks=[0, 128, 255])


def gen():
    while stream.is_active():
        data = stream.read(chunk)
        data_int = struct.unpack(str(chunk*2) + 'B', data)
        data_np = np.array(data_int, dtype='b')[::2] + 128
        yf = np.fft.fft(data_int)
        yield data_np, yf


def init():
    lf.set_ydata(np.random.rand(chunk))
    return lf,


def update(data):
    ax1.figure.canvas.draw()
    ax2.figure.canvas.draw()
    l.set_ydata(data[0])
    lf.set_ydata(np.abs(data[1][:chunk]) / (128 * chunk))
    return lf,


animate = FuncAnimation(fig, update, gen, blit=True, interval=0, repeat=False, init_func=init)
plt.show()
stream.stop_stream()
stream.close()