marc nicole <mk1853387@gmail.com> wrote or quoted:

I hope this question is not very far from the main topic of this list, but
I have a hard time finding a way to check whether audio data samples are >containing empty noise or actual significant voice/noise.

The Spectral Flatness Measure (SFM), also called Wiener entropy, can
separate the wheat from the chaff when it comes to how noise-like
a signal is. This measure runs the gamut from 0 to 1, where:
1 means you've hit pay dirt with perfect white noise (flat spectrum),
0 is as pure as a Napa Valley Chardonnay (single frequency).
(Everything in between is just different shades of gnarly.)

import numpy as np
from scipy.signal import welch

def noiseness(signal, fs):
# Compute the power spectral density
f, psd = welch(signal, fs, nperseg=min(len(signal), 256))

# Compute geometric mean of PSD
geometric_mean = np.exp(np.mean(np.log(psd + 1e-10)))

# Compute arithmetic mean of PSD
arithmetic_mean = np.mean(psd)

# Calculate Spectral Flatness Measure
sfm = geometric_mean / arithmetic_mean

return sfm

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)

ram@zedat.fu-berlin.de (Stefan Ram) wrote or quoted:

marc nicole <mk1853387@gmail.com> wrote or quoted:

I hope this question is not very far from the main topic of this list, but >>I have a hard time finding a way to check whether audio data samples are >>containing empty noise or actual significant voice/noise.

The Spectral Flatness Measure (SFM), also called Wiener entropy, can
separate the wheat from the chaff when it comes to how noise-like
a signal is.

You can also peep the envelope flatness (the flatness of the
volume). If you've got some white noise that's not bringing much to
the table, that envelope should be flatter than a pancake at IHOP.

import librosa
import numpy as np

def measure_volume_flatness(audio_path, sr=None):
# Load the audio file
y, sr = librosa.load(audio_path, sr=sr)

# Calculate the root mean square (RMS) energy for each frame
frame_length = 2048
hop_length = 512
rms = librosa.feature.rms(y=y, frame_length=frame_length, hop_length=hop_length)[0]

# Calculate the dynamic range
db_range = librosa.amplitude_to_db(np.max(rms)) - librosa.amplitude_to_db(np.min(rms))

# Normalize the dynamic range to a 0-1 scale
# Assuming a maximum possible dynamic range of 120 dB
flatness = 1 - (db_range / 120)

return flatness

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)