Develop Reference

How to Encode a 16 Bit WAV file in 8 bit in Python?

I am trying to play a sound from a sawtooth wave. I created the waveform in Python and was able to save it as a WAV file, but when I try to play it it says the file is unplayable because the file type is unsupported, the file extension is incorrect, or the file is corrupt. I used this individual's tutorial (https://thehackerdiary.wordpress.com/2017/06/09/it-is-ridiculously-easy-to-generate-any-audio-signal-using-python/) and they worked around it by encoding the raw waveform from 16 bit to 8 bit in Audacity. How can this be done using only Python?
import soundfile
data, samplerate = soundfile.read('sawtooth_100_hz.wav')
soundfile.write('sawtooth_100_hz_8bit.wav', data, samplerate, subtype='PCM_S8')
^^ I tried this and got the following error: ValueError: Invalid combination of format, subtype and endian

I think the people who wrote this tutorial went for the long way. There is an easier way to convert a NumPy array to a wav file which is used below to generate the same wav file as the one generated in the tutorial:
import numpy as np
from scipy.io import wavfile
sampling_rate = 44100
freq = 440
samples = 44100
x = np.arange(samples)
y = 100*np.sin(2 * np.pi * freq * x / sampling_rate)
wavfile.write("test.wav", sampling_rate, y)
And you can use wavfile.read() method to read this file with no problem

Surprisingly, the underlying libsndfile library doesn't support WAV files with signed 8 bit samples (only unsigned), see http://www.mega-nerd.com/libsndfile/#Features.
You can also check this with the soundfile module:
>>> import soundfile as sf
>>> sf.available_subtypes('wav')
{'PCM_16': 'Signed 16 bit PCM', 'PCM_24': 'Signed 24 bit PCM', 'PCM_32': 'Signed 32 bit PCM', 'PCM_U8': 'Unsigned 8 bit PCM', 'FLOAT': '32 bit float', 'DOUBLE': '64 bit float', 'ULAW': 'U-Law', 'ALAW': 'A-Law', 'IMA_ADPCM': 'IMA ADPCM', 'MS_ADPCM': 'Microsoft ADPCM', 'GSM610': 'GSM 6.10', 'G721_32': '32kbs G721 ADPCM'}
You could try using AIFF or FLAC instead?
Or you could create a RAW file (i.e. a headerless file containing no information about its own data format), which is incidentally what they did in the tutorial you were mentioning (note that they are using these options: -t raw -e signed -b 8).
For a bit more information about creating and playing signals, see:
https://nbviewer.jupyter.org/github/mgeier/python-audio/blob/master/simple-signals.ipynb
https://nbviewer.jupyter.org/github/spatialaudio/communication-acoustics-exercises/blob/master/intro.ipynb

It sounds like you just want to generate the samples and playback from within Python?
If so, it looks like library "sounddevice" will let you write samples directly to your audio device:
https://python-sounddevice.readthedocs.io/en/0.3.15/usage.html#playback
I'm not in a python environment right now, so haven't tested, but mixing it with your sample code would just be:
import sounddevice as sd
import numpy as np
sampling_rate = 44100
freq = 440
samples = 44100
x = np.arange(samples)
y = 100*np.sin(2 * np.pi * freq * x / sampling_rate)
sd.play(y, sampling_rate)
Sounddevice's author is on SO, see his reply to a similar question: https://stackoverflow.com/a/34179010/1339735
You might have some scaling to do - not sure if it accepts values from -1 to 1 like most float playback or +/- 100 like in your example.

All of the above answers were helpful, but ultimately I found a solution to my problem from this thread: How to generate audio from a numpy array?
This was my code:
import numpy as np
from scipy.io.wavfile import write
from scipy import signal as sg
#data = np.random.uniform(-1,1,44100) # 44100 random samples between -1 and 1
sampling_rate = 44100 ## Sampling Rate
freq = 150 ## Frequency (in Hz)
duration = 3 # in seconds, may be float
t = np.linspace(0, duration, sampling_rate*duration) # Creating time vector
data = sg.sawtooth(2 * np.pi * freq * t, 0) # Sawtooth signal
'''
Scaling data to 16 bit. Divide each number by max number in array to get
fraction and multiply data by 32767 because that is the max value a 16 bit
integer can take
'''
scaled = np.int16(data/np.max(np.abs(data)) * 32767)
write('test.wav', 44100, scaled) # Write to file. Can be overridden

NumPy Fast Fourier transform (FFT) does not work on sine wave generated in Audacity

I am trying to use the NumPy library for Python to do some frequency analysis. I have two .wav files that both contain a 440 Hz sine wave. One of them I generated with the NumPy sine function, and the other I generated in Audacity. The FFT works on the Python-generated one, but does nothing on the Audacity one.
Here are links to the two files:
The non-working file: 440_audacity.wav
The working file: 440_gen.wav
This is the code I am using to do the Fourier transform:
import numpy as np
import matplotlib.pyplot as plt
import scipy.io.wavfile as wave
infile = "440_gen.wav"
rate, data = wave.read(infile)
data = np.array(data)
data_fft = np.fft.fft(data)
frequencies = np.abs(data_fft)
plt.subplot(2,1,1)
plt.plot(data[:800])
plt.title("Original wave: " + infile)
plt.subplot(2,1,2)
plt.plot(frequencies)
plt.title("Fourier transform results")
plt.xlim(0, 1000)
plt.tight_layout()
plt.show()
I have two 16-bit PCM .wav files, one from Audacity and one created with the NumPy sine function. The NumPy-generated one gives the following (correct) result, with the spike at 440Hz:
The one I created with Audacity, although the waveform appears identical, does not give any result on the Fourier transform:
I admit I am at a loss here. The two files should contain in effect the same data. They are encoded the same way, and the wave forms appear identical on the upper graph.
Here is the code used to generate the working file:
import numpy as np
import wave
import struct
import matplotlib.pyplot as plt
from operator import add
freq_one = 440.0
num_samples = 44100
sample_rate = 44100.0
amplitude = 12800
file = "440_gen.wav"
s1 = [np.sin(2 * np.pi * freq_one * x/sample_rate) * amplitude for x in range(num_samples)]
sine_one = np.array(s1)
nframes = num_samples
comptype = "NONE"
compname="not compressed"
nchannels = 1
sampwidth = 2
wav_file = wave.open(file, 'w')
wav_file.setparams((nchannels, sampwidth, int(sample_rate), nframes, comptype, compname))
for s in sine_one:
wav_file.writeframes(struct.pack('h', int(s)))

Let me explain why your code doesn't work. And why it works with [:44100].
First of all, you have different files:
440_gen.wav = 1 sec and 44100 samples (counts)
440_audacity.wav = 5 sec and 220500 samples (counts)
Since for 440_gen.wav in FFT you use the number of reference points N=44100 and the sample rate 44100, your frequency resolution is 1 Hz (bins are followed in 1 Hz increments).
Therefore, on the graph, each FFT sample corresponds to a delta equal to 1 Hz.
plt.xlim(0, 1000) just corresponds to the range 0-1000 Hz.
However, for 440_audacity.wav in FFT, you use the number of reference points N=220500 and the sample rate 44100. Your frequency resolution is 0.2 Hz (bins follow in 0.2 Hz increments) - on the graph, each FFT sample corresponds to a frequency in 0.2 Hz increments (min-max = +(-) 22500 Hz).
plt.xlim(0, 1000) just corresponds to the range 1000x0.2 = 0-200 Hz.
That is why the result is not visible - it does not fall within this range.
plt.xlim (0, 5000) will correct your situation and extend the range to 0-1000 Hz.
The solution [:44100] that jwalton brought in really only forces the FFT to use N = 44100. And this repeats the situation with the calculation for 440_gen.wav
A more correct solution to your problem is to use the N (Windows Size) parameter in the code and the np.fft.fftfreq() function.
Sample code below.
I also recommend an excellent article https://realpython.com/python-scipy-fft/
import numpy as np
import matplotlib.pyplot as plt
import scipy.io.wavfile as wave
N = 44100 # added
infile = "440_audacity.wav"
rate, data = wave.read(infile)
data = np.array(data)
data_fft = np.fft.fft(data, N) # added N
frequencies = np.abs(data_fft)
x_freq = np.fft.fftfreq(N, 1/44100) # added
plt.subplot(2,1,1)
plt.plot(data[:800])
plt.title("Original wave: " + infile)
plt.subplot(2,1,2)
plt.plot(x_freq, frequencies) # added x_freq
plt.title("Fourier transform results")
plt.xlim(0, 1000)
plt.tight_layout()
plt.show()

Since answering this question #Konyukh Fyodorov was able to provide a better and properly justified solution (below).
The following worked for me and produced the plots as expected. Unfortunately I cannot piece together quite why this works, but I'm sharing this solution in the hope it may assist someone else to make that leap.
import numpy as np
import matplotlib.pyplot as plt
import scipy.io.wavfile as wave
infile = "440_gen.wav"
rate, data = wave.read(infile)
data = np.array(data)
# Use first 44100 datapoints in transform
data_fft = np.fft.fft(data[:44100])
frequencies = np.abs(data_fft)
plt.subplot(2,1,1)
plt.plot(data[:800])
plt.title("Original wave: " + infile)
plt.subplot(2,1,2)
plt.plot(frequencies)
plt.title("Fourier transform results")
plt.xlim(0, 1000)
plt.tight_layout()
plt.show()

FFT spectrogram in python

I have python 3.4.
I transmitted a 2MHz (for example) frequency and received the cavitation over the time (until I stopped the measurement).
I want to get a spectrogram (cavitation vs frequency) and more interesting is a spectrogram of cavitation over the time of the sub-harmonic (1MHz) frequency.
The data is saved in sdataA (=cavitation), and t (=measurement time)
I tried to save fft in FFTA
FFTA = np.array([])
FFTA = np.fft.fft(dataA)
FFTA = np.append(FFTA, dataA)
I got real and complex numbers
Then I took only half (from 0 to 1MHz) and save the real and complex data.
nA = int(len(FFTA)/2)
yAre = FFTA[range(nA)].real
yAim = FFTA[range(nA)].imag
I tried to get the frequencies by:
FFTAfreqs = np.fft.fftfreq(len(yAre))
But it is totally wrong (I printed the data by print (FFTAfreqs))
I also plotted the data and again it's wrong:
plt.plot(t, FFTA[range(n)].real, 'b-', t, FFTA[range(n)].imag, 'r--')
plt.legend(('real', 'imaginary'))
plt.show()
How can I output a spectrogram of cavitation over the time of the sub-harmonic (1MHz) frequency?
EDIT:
Data example:
see a sample of 'dataA' and 'time':
dataA = [6.08E-04,2.78E-04,3.64E-04,3.64E-04,4.37E-04,4.09E-04,4.49E-04,4.09E-04,3.52E-04,3.24E-04,3.92E-04,3.24E-04,2.67E-04,3.24E-04,2.95E-04,2.95E-04,4.94E-04,4.09E-04,3.64E-04,3.07E-04]
time = [0.00E+00,4.96E-07,9.92E-07,1.49E-06,1.98E-06,2.48E-06,2.98E-06,3.47E-06,3.97E-06,4.46E-06,4.96E-06,5.46E-06,5.95E-06,6.45E-06,6.94E-06,7.44E-06,7.94E-06,8.43E-06,8.93E-06,9.42E-06]
EDIT II:
From #Martin example I tried the following code, please let me know if I did it right.
In the case that dataA and Time are saved as h5 files (or the data that I posted already)
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd
dfdata = pd.read_hdf("C:\\data_python\\DataA.h5")
dft = pd.read_hdf("C:\\data_python\\time.h5")
dft_cor = int((len(dft)-2)*4.96E-6) # calculating the measured time
fs = 2000000 #sampling frequency 2MHz
CHUNK = 10000
signal_time = dft_cor # seconds
def sine(freq,fs,secs):
data=dfdata
wave = np.sin(freq*2*np.pi*data)
return wave
a1 = sine(fs,fs,120)
a2 = sine(fs/2,fs,120)
signal = a1+a2
afft = np.abs(np.fft.fft(signal[0:CHUNK]))
freqs = np.linspace(0,fs,CHUNK)[0:int(fs/2)]
spectrogram_chunk = freqs/np.amax(freqs*1.0)
# Plot spectral analysis
plt.plot(freqs[0:1000000],afft[0:1000000]) # 0-1MHz
plt.show()
number_of_chunks = 1000
# Empty spectrogram
Spectrogram = np.zeros(shape = [CHUNK,number_of_chunks])
for i in range(number_of_chunks):
afft = np.abs(np.fft.fft(signal[i*CHUNK:(1+i)*CHUNK]))
freqs = np.linspace(0,fs,CHUNK)[0:int(fs/2)]
spectrogram_chunk = afft/np.amax(afft*1.0)
try:
Spectrogram[:,i]=spectrogram_chunk
except:
break
import cv2
Spectrogram = Spectrogram[0:1000000,:]
cv2.imshow('spectrogram',np.uint8(255*Spectrogram/np.amax(Spectrogram)))
cv2.waitKey()
cv2.destroyAllWindows()

It seems your problem is not in Python but in understanding what is Spectrogram.
Spectrogram is sequences of spectral analysis of a signal.
1) You need to cut your signal in CHUNKS.
2) Do spectral analysis of these CHUNKS and stick it together.
Example:
You have 1 second of audio recoding (44100 HZ sampling). That means the recording will have 1s * 44100 -> 44100 samples. You define CHUNK size = 1024 (for example).
For each chunk you will do FFT, and stick it together into 2D matrix (X axis - FFT of the CHUNK, Y axis - CHUNK number,). 44100 samples / CHUNK ~ 44 FFTs, each of the FFT covers 1024/44100~0.023 seconds of the signal
The bigger the CHUNK, the more accurate Spectrogram is, but less 'realtime'.
The smaller the CHUNK is, the less acurate is the Spectrogram, but you have more measurements as you measure frequencies 'more often'.
If you need 1MHZ - actually you cannot use anything higher than 1MHZ, you just take half of the resulting FFT array - and it doesnt matter which half, because 1MHZ is just the half of your sampling frequency, and the FFT is mirroring anything that is higher than 1/2 of sampling frequency.
About FFT, you dont want complex numbers. You want to do
FFT = np.abs(FFT) # Edit - I just noticed you use '.real', but I will keep it here
because you want real numbers.
Preparation for Spectrogram - example of Spectrogram
Audio Signal with 150HZ wave and 300HZ Wave
import numpy as np
import matplotlib.pyplot as plt
fs = 44100#sampling frequency
CHUNK = 10000
signal_time = 20 # seconds
def sine(freq,fs,secs):
data=np.arange(fs*secs)/(fs*1.0)
wave = np.sin(freq*2*np.pi*data)
return wave
a1 = sine(150,fs,120)
a2 = sine(300,fs,120)
signal = a1+a2
afft = np.abs(np.fft.fft(signal[0:CHUNK]))
freqs = np.linspace(0,fs,CHUNK)[0:int(fs/2)]
spectrogram_chunk = freqs/np.amax(freqs*1.0)
# Plot spectral analysis
plt.plot(freqs[0:250],afft[0:250])
plt.show()
number_of_chunks = 1000
# Empty spectrogram
Spectrogram = np.zeros(shape = [CHUNK,number_of_chunks])
for i in range(number_of_chunks):
afft = np.abs(np.fft.fft(signal[i*CHUNK:(1+i)*CHUNK]))
freqs = np.linspace(0,fs,CHUNK)[0:int(fs/2)]
#plt.plot(spectrogram_chunk[0:250],afft[0:250])
#plt.show()
spectrogram_chunk = afft/np.amax(afft*1.0)
#print(signal[i*CHUNK:(1+i)*CHUNK].shape)
try:
Spectrogram[:,i]=spectrogram_chunk
except:
break
import cv2
Spectrogram = Spectrogram[0:250,:]
cv2.imshow('spectrogram',np.uint8(255*Spectrogram/np.amax(Spectrogram)))
cv2.waitKey()
cv2.destroyAllWindows()
Spectral analysis of single CHUNK
Spectrogram

FFT of data received from PyAudio gives wrong frequency

My main task is to recognize a human humming from a microphone in real time. As the first step to recognizing signals in general, I have made a 5 seconds recording of a 440 Hz signal generated from an app on my phone and tried to detect the same frequency.
I used Audacity to plot and verify the spectrum from the same 440Hz wav file and I got this, which shows that 440Hz is indeed the dominant frequency :
(https://i.imgur.com/2UImEkR.png)
To do this with python, I use the PyAudio library and refer this blog. The code I have so far which I run with the wav file is this :
"""PyAudio Example: Play a WAVE file."""
import pyaudio
import wave
import sys
import struct
import numpy as np
import matplotlib.pyplot as plt
CHUNK = 1024
if len(sys.argv) < 2:
print("Plays a wave file.\n\nUsage: %s filename.wav" % sys.argv[0])
sys.exit(-1)
wf = wave.open(sys.argv[1], 'rb')
p = pyaudio.PyAudio()
stream = p.open(format=p.get_format_from_width(wf.getsampwidth()),
channels=wf.getnchannels(),
rate=wf.getframerate(),
output=True)
data = wf.readframes(CHUNK)
i = 0
while data != '':
i += 1
data_unpacked = struct.unpack('{n}h'.format(n= len(data)/2 ), data)
data_np = np.array(data_unpacked)
data_fft = np.fft.fft(data_np)
data_freq = np.abs(data_fft)/len(data_fft) # Dividing by length to normalize the amplitude as per https://www.mathworks.com/matlabcentral/answers/162846-amplitude-of-signal-after-fft-operation
print("Chunk: {} max_freq: {}".format(i,np.argmax(data_freq)))
fig = plt.figure()
ax = fig.add_subplot(1,1,1)
ax.plot(data_freq)
ax.set_xscale('log')
plt.show()
stream.write(data)
data = wf.readframes(CHUNK)
stream.stop_stream()
stream.close()
p.terminate()
In the output, I get that the max frequency is 10 for all the chunks and an example of one of the plots is :
(https://i.imgur.com/zsAXME5.png)
I had expected this value to be 440 instead of 10 for all the chunks. I admit I know very little about the theory of FFTs and I appreciate any help in letting my solve this.
EDIT:
The sampling rate is 44100. no. of channels is 2 and sample width is also 2.

Forewords
As xdurch0 pointed out, you are reading a kind of index instead of a frequency. If you are about to make all computation by yourself you need to compute you own frequency vector before plotting if you want to get consistent result. Reading this answer may help you towards the solution.
The frequency vector for FFT (half plane) is:
f = np.linspace(0, rate/2, N_fft/2)
Or (full plane):
f = np.linspace(-rate/2, rate/2, N_fft)
On the other hand we can delegate most of the work to the excellent scipy.signal toolbox which aims to cope with this kind of problems (and many more).
MCVE
Using scipy package it is straight forward to get the desired result for a simple WAV file with a single frequency (source):
import numpy as np
from scipy import signal
from scipy.io import wavfile
import matplotlib.pyplot as plt
# Read the file (rate and data):
rate, data = wavfile.read('tone.wav') # See source
# Compute PSD:
f, P = signal.periodogram(data, rate) # Frequencies and PSD
# Display PSD:
fig, axe = plt.subplots()
axe.semilogy(f, P)
axe.set_xlim([0,500])
axe.set_ylim([1e-8, 1e10])
axe.set_xlabel(r'Frequency, $\nu$ $[\mathrm{Hz}]$')
axe.set_ylabel(r'PSD, $P$ $[\mathrm{AU^2Hz}^{-1}]$')
axe.set_title('Periodogram')
axe.grid(which='both')
Basically:
Read the wav file and get the sample rate (here 44.1kHz);
Compute the Power Spectrum Density and frequencies;
Then display it with matplotlib.
This outputs:
Find Peak
Then we can find the frequency of the first highest peak (P>1e-2, this criterion is subject to tuning) using find_peaks:
idx = signal.find_peaks(P, height=1e-2)[0][0]
f[idx] # 440.0 Hz
Putting all together it merely boils down to:
def freq(filename, setup={'height': 1e-2}):
rate, data = wavfile.read(filename)
f, P = signal.periodogram(data, rate)
return f[signal.find_peaks(P, **setup)[0][0]]
Handling multiple channels
I tried this code with my wav file, and got the error for the line
axe.semilogy(f, Pxx_den) as follows : ValueError: x and y must have
same first dimension. I checked the shapes and f has (2,) while
Pxx_den has (220160,2). Also, the Pxx_den array seems to have all
zeros only.
Wav file can hold multiple channels, mainly there are mono or stereo files (max. 2**16 - 1 channels). The problem you underlined occurs because of multiple channels file (stereo sample).
rate, data = wavfile.read('aaaah.wav') # Shape: (46447, 2), Rate: 48 kHz
It is not well documented, but the method signal.periodogram also performs on matrix and its input is not directly consistent with wavfile.read output (they perform on different axis by default). So we need to carefully orient dimensions (using axis switch) when performing PSD:
f, P = signal.periodogram(data, rate, axis=0, detrend='linear')
It also works with Transposition data.T but then we need to back transpose the result.
Specifying the axis solve the issue: frequency vector is correct and PSD is not null everywhere (before it performed on the axis=1 which is of length 2, in your case it performed 220160 PSD on 2-samples signals we wanted the converse).
The detrend switch ensure the signal has zero mean and its linear trend is removed.
Real application
This approach should work for real chunked samples, provided chunks hold enough data (see Nyquist-Shannon sampling theorem). Then data are sub-samples of the signal (chunks) and rate is kept constant since it does not change during the process.
Having chunks of size 2**10 seems to work, we can identify specific frequencies from them:
f, P = signal.periodogram(data[:2**10,:], rate, axis=0, detrend='linear') # Shapes: (513,) (513, 2)
idx0 = signal.find_peaks(P[:,0], threshold=0.01, distance=50)[0] # Peaks: [46.875, 2625., 13312.5, 16921.875] Hz
fig, axe = plt.subplots(2, 1, sharex=True, sharey=True)
axe[0].loglog(f, P[:,0])
axe[0].loglog(f[idx0], P[idx0,0], '.')
# [...]
At this point, the trickiest part is the fine tuning of find-peaks method to catch desired frequencies. You may need to consider to pre-filter your signal or post-process the PSD in order to make the identification easier.

Get frequency value (hz) in split wav audio file by python

I have simple case, i'm recording bee sound in Bee have by python pyAudio. After recording i have to split this record in 10 sec chunks and analyze this chunks by python wave, numpy or python scipy, numpy, i don't know what is easiest way.
I would like to read the record then split it to 10 sec chunk and apply fft or rfft and after that i need get dominant frequencies in Hz in chunk.
I collect this values with timestamp in histogram for histogram plot.
Right now i have some examples where i can get whole record and make plot by python matplotlib, scipy.signal but i don't know how to separated it into listo of Hz values.
If i have completely wrong way please tell me. Thx for advice.
import numpy as np
import struct, wave
def main():
audio = wave.open('wav_data/18-08-07_09_10_12.wav', 'rb')
rate = audio.getframerate()
num_frames = audio.getnframes()
dur = int(num_frames / rate)
fmt = "%ih" % rate
for x in range(dur):
data = audio.readframes(rate)
data_int = struct.unpack(fmt, data)
data_np = np.array(data_int, dtype='b')
w = np.fft.rfft(data_np)
freqs = np.fft.fftfreq(len(w))
idx = np.argmax(w)
freq = freqs[idx]
freq_in_hz = abs(freq * rate)
print(freq_in_hz)
if __name__ == "__main__":
main()