Test projects : background-noise-reduce

Test_noise_reduction/noise_reduction-master/README.md

+# noise_reduction
+
+> See test results on: [https://dodiku.github.io/noise_reduction/](https://dodiku.github.io/noise_reduction/)
+
+## Audio enhancements feature tests in Python3
+
+#### Installation
+To install:
+1. ``$ brew install sox``
+1. ``$ brew install vorbis-tools``
+1. Create a virtualenv
+1. Install dependencies in one of two options:
+  - manually *(recommended)*:  
+      ``$ pip3 install librosa``  
+      ``$ pip3 install pysndfx``
+
+  - or automatically using pip:  
+      ``$ pip3 install -r requirements.txt``
+
+To run:  
+``$ python3 noise.py``
+
+
+#### Interesting resources:
+- LibROSA ([documentation](http://librosa.github.io/librosa/index.html) + [repository](https://github.com/librosa/librosa) + [paper](https://bmcfee.github.io/papers/scipy2015_librosa.pdf))
+- Think DSP ([book](http://greenteapress.com/wp/think-dsp/) + [repository](https://github.com/AllenDowney/ThinkDSP/))
+- Pyo ([blog post](http://www.matthieuamiguet.ch/blog/diy-guitar-effects-python) + [repository](https://github.com/belangeo/pyo))
+- pysndfx ([repository](https://github.com/carlthome/python-audio-effects/tree/04dbee6063b0537b63346bb1e55deb03406e1170/pysndfx))
+
+#### A bit less relevant papers:
+- Noise Cancellation Method for Robust Speech Recognition ([PDF](http://research.ijcaonline.org/volume45/number11/pxc3879438.pdf))
+- Robust Features for Noisy Speech Recognition using MFCC Computation from Magnitude Spectrum of Higher Order Autocorrelation Coefficients
+([PDF](https://pdfs.semanticscholar.org/a483/5f28c02f07e6bef04ff9db948505dc990af7.pdf))
+- Improving the Noise-Robustness of Mel-Frequency Cepstral Coefficients for Speech Processing
+([PDF](http://www.sapaworkshops.org/2006/2006/papers/131.pdf))

Test_noise_reduction/noise_reduction-master/matplotlibrc

+backend: TkAgg

Test_noise_reduction/noise_reduction-master/noise.py

+import librosa
+from pysndfx import AudioEffectsChain
+import numpy as np
+import math
+import python_speech_features
+import scipy as sp
+from scipy import signal
+import soundfile
+
+# http://python-speech-features.readthedocs.io/en/latest/
+# https://github.com/jameslyons/python_speech_features
+# http://practicalcryptography.com/miscellaneous/machine-learning/guide-mel-frequency-cepstral-coefficients-mfccs/#deltas-and-delta-deltas
+
+
+# http://dsp.stackexchange.com/search?q=noise+reduction/
+
+'''------------------------------------
+FILE READER:
+    receives filename,
+    returns audio time series (y) and sampling rate of y (sr)
+------------------------------------'''
+def read_file(file_name):
+    sample_file = file_name
+    sample_directory = '00_samples/'
+    sample_path = sample_directory + sample_file
+
+    # generating audio time series and a sampling rate (int)
+    y, sr = librosa.load(sample_path)
+
+    return y, sr
+
+'''------------------------------------
+NOISE REDUCTION USING POWER:
+    receives an audio matrix,
+    returns the matrix after gain reduction on noise
+------------------------------------'''
+def reduce_noise_power(y, sr):
+
+    cent = librosa.feature.spectral_centroid(y=y, sr=sr)
+
+    threshold_h = round(np.median(cent))*1.5
+    threshold_l = round(np.median(cent))*0.1
+
+    less_noise = AudioEffectsChain().lowshelf(gain=-30.0, frequency=threshold_l, slope=0.8).highshelf(gain=-12.0, frequency=threshold_h, slope=0.5)#.limiter(gain=6.0)
+    y_clean = less_noise(y)
+
+    return y_clean
+
+
+'''------------------------------------
+NOISE REDUCTION USING CENTROID ANALYSIS:
+    receives an audio matrix,
+    returns the matrix after gain reduction on noise
+------------------------------------'''
+
+def reduce_noise_centroid_s(y, sr):
+
+    cent = librosa.feature.spectral_centroid(y=y, sr=sr)
+
+    threshold_h = np.max(cent)
+    threshold_l = np.min(cent)
+
+    less_noise = AudioEffectsChain().lowshelf(gain=-12.0, frequency=threshold_l, slope=0.5).highshelf(gain=-12.0, frequency=threshold_h, slope=0.5).limiter(gain=6.0)
+
+    y_cleaned = less_noise(y)
+
+    return y_cleaned
+
+def reduce_noise_centroid_mb(y, sr):
+
+    cent = librosa.feature.spectral_centroid(y=y, sr=sr)
+
+    threshold_h = np.max(cent)
+    threshold_l = np.min(cent)
+
+    less_noise = AudioEffectsChain().lowshelf(gain=-30.0, frequency=threshold_l, slope=0.5).highshelf(gain=-30.0, frequency=threshold_h, slope=0.5).limiter(gain=10.0)
+    # less_noise = AudioEffectsChain().lowpass(frequency=threshold_h).highpass(frequency=threshold_l)
+    y_cleaned = less_noise(y)
+
+
+    cent_cleaned = librosa.feature.spectral_centroid(y=y_cleaned, sr=sr)
+    columns, rows = cent_cleaned.shape
+    boost_h = math.floor(rows/3*2)
+    boost_l = math.floor(rows/6)
+    boost = math.floor(rows/3)
+
+    # boost_bass = AudioEffectsChain().lowshelf(gain=20.0, frequency=boost, slope=0.8)
+    boost_bass = AudioEffectsChain().lowshelf(gain=16.0, frequency=boost_h, slope=0.5)#.lowshelf(gain=-20.0, frequency=boost_l, slope=0.8)
+    y_clean_boosted = boost_bass(y_cleaned)
+
+    return y_clean_boosted
+
+
+'''------------------------------------
+NOISE REDUCTION USING MFCC:
+    receives an audio matrix,
+    returns the matrix after gain reduction on noise
+------------------------------------'''
+def reduce_noise_mfcc_down(y, sr):
+
+    hop_length = 512
+
+    ## librosa
+    # mfcc = librosa.feature.mfcc(y=y, sr=sr, hop_length=hop_length, n_mfcc=13)
+    # librosa.mel_to_hz(mfcc)
+
+    ## mfcc
+    mfcc = python_speech_features.base.mfcc(y)
+    mfcc = python_speech_features.base.logfbank(y)
+    mfcc = python_speech_features.base.lifter(mfcc)
+
+    sum_of_squares = []
+    index = -1
+    for r in mfcc:
+        sum_of_squares.append(0)
+        index = index + 1
+        for n in r:
+            sum_of_squares[index] = sum_of_squares[index] + n**2
+
+    strongest_frame = sum_of_squares.index(max(sum_of_squares))
+    hz = python_speech_features.base.mel2hz(mfcc[strongest_frame])
+
+    max_hz = max(hz)
+    min_hz = min(hz)
+
+    speech_booster = AudioEffectsChain().highshelf(frequency=min_hz*(-1)*1.2, gain=-12.0, slope=0.6).limiter(gain=8.0)
+    y_speach_boosted = speech_booster(y)
+
+    return (y_speach_boosted)
+
+def reduce_noise_mfcc_up(y, sr):
+
+    hop_length = 512
+
+    ## librosa
+    # mfcc = librosa.feature.mfcc(y=y, sr=sr, hop_length=hop_length, n_mfcc=13)
+    # librosa.mel_to_hz(mfcc)
+
+    ## mfcc
+    mfcc = python_speech_features.base.mfcc(y)
+    mfcc = python_speech_features.base.logfbank(y)
+    mfcc = python_speech_features.base.lifter(mfcc)
+
+    sum_of_squares = []
+    index = -1
+    for r in mfcc:
+        sum_of_squares.append(0)
+        index = index + 1
+        for n in r:
+            sum_of_squares[index] = sum_of_squares[index] + n**2
+
+    strongest_frame = sum_of_squares.index(max(sum_of_squares))
+    hz = python_speech_features.base.mel2hz(mfcc[strongest_frame])
+
+    max_hz = max(hz)
+    min_hz = min(hz)
+
+    speech_booster = AudioEffectsChain().lowshelf(frequency=min_hz*(-1), gain=12.0, slope=0.5)#.highshelf(frequency=min_hz*(-1)*1.2, gain=-12.0, slope=0.5)#.limiter(gain=8.0)
+    y_speach_boosted = speech_booster(y)
+
+    return (y_speach_boosted)
+
+'''------------------------------------
+NOISE REDUCTION USING MEDIAN:
+    receives an audio matrix,
+    returns the matrix after gain reduction on noise
+------------------------------------'''
+
+def reduce_noise_median(y, sr):
+    y = sp.signal.medfilt(y,3)
+    return (y)
+
+
+'''------------------------------------
+SILENCE TRIMMER:
+    receives an audio matrix,
+    returns an audio matrix with less silence and the amout of time that was trimmed
+------------------------------------'''
+def trim_silence(y):
+    y_trimmed, index = librosa.effects.trim(y, top_db=20, frame_length=2, hop_length=500)
+    trimmed_length = librosa.get_duration(y) - librosa.get_duration(y_trimmed)
+
+    return y_trimmed, trimmed_length
+
+
+'''------------------------------------
+AUDIO ENHANCER:
+    receives an audio matrix,
+    returns the same matrix after audio manipulation
+------------------------------------'''
+def enhance(y):
+    apply_audio_effects = AudioEffectsChain().lowshelf(gain=10.0, frequency=260, slope=0.1).reverb(reverberance=25, hf_damping=5, room_scale=5, stereo_depth=50, pre_delay=20, wet_gain=0, wet_only=False)#.normalize()
+    y_enhanced = apply_audio_effects(y)
+
+    return y_enhanced
+
+'''------------------------------------
+OUTPUT GENERATOR:
+    receives a destination path, file name, audio matrix, and sample rate,
+    generates a wav file based on input
+------------------------------------'''
+def output_file(destination ,filename, y, sr, ext=""):
+    destination = destination + filename[:-4] + ext + '.wav'
+    librosa.output.write_wav(destination, y, sr)
+
+
+'''------------------------------------
+LOGIC:
+    [1] load file
+    [2] reduce noise
+    [3] trim silence
+    [4] output file
+
+sample files:
+    01_counting.m4a
+    02_wind_and_cars.m4a
+    03_truck.m4a
+    04_voices.m4a
+    05_ambeint.m4a
+    06_office.m4a
+------------------------------------'''
+samples = ['01_counting.m4a','02_wind_and_cars.m4a','03_truck.m4a','04_voices.m4a','05_ambeint.m4a','06_office.m4a']
+
+for s in samples:
+    # reading a file
+    filename = s
+    y, sr = read_file(filename)
+
+    # reducing noise using db power
+    y_reduced_power = reduce_noise_power(y, sr)
+    y_reduced_centroid_s = reduce_noise_centroid_s(y, sr)
+    y_reduced_centroid_mb = reduce_noise_centroid_mb(y, sr)
+    y_reduced_mfcc_up = reduce_noise_mfcc_up(y, sr)
+    y_reduced_mfcc_down = reduce_noise_mfcc_down(y, sr)
+    y_reduced_median = reduce_noise_median(y, sr)
+
+    # trimming silences
+    y_reduced_power, time_trimmed = trim_silence(y_reduced_power)
+    # print (time_trimmed)
+
+    y_reduced_centroid_s, time_trimmed = trim_silence(y_reduced_centroid_s)
+    # print (time_trimmed)
+
+    y_reduced_power, time_trimmed = trim_silence(y_reduced_power)
+    # print (time_trimmed)
+
+    y_reduced_centroid_mb, time_trimmed = trim_silence(y_reduced_centroid_mb)
+    # print (time_trimmed)
+
+    y_reduced_mfcc_up, time_trimmed = trim_silence(y_reduced_mfcc_up)
+    # print (time_trimmed)
+
+    y_reduced_mfcc_down, time_trimmed = trim_silence(y_reduced_mfcc_down)
+    # print (time_trimmed)
+
+    y_reduced_median, time_trimmed = trim_silence(y_reduced_median)
+
+    # generating output file [1]
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y_reduced_power, sr, '_pwr')
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y_reduced_centroid_s, sr, '_ctr_s')
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y_reduced_centroid_mb, sr, '_ctr_mb')
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y_reduced_mfcc_up, sr, '_mfcc_up')
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y_reduced_mfcc_down, sr, '_mfcc_down')
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y_reduced_median, sr, '_median')
+    output_file('01_samples_trimmed_noise_reduced/' ,filename, y, sr, '_org')

Test_noise_reduction/noise_reduction-master/requirements.txt

+appdirs==1.4.3
+audioread==2.1.4
+Cython==0.25.2
+decorator==4.0.11
+joblib==0.11
+librosa==0.5.0
+numpy==1.12.1
+packaging==16.8
+pyparsing==2.2.0
+pysndfx==0.1.0
+python-speech-features==0.5
+resampy==0.1.5
+scikit-learn==0.18.1
+scipy==0.19.0
+six==1.10.0

Test_noise_reduction/noise_reduction-master/styles.css

+html {
+  background-color: LightSeaGreen;
+}
+
+body {
+  font-family: "Courier New", Courier, monospace;
+  color: #222;
+  font-size: 0.9em;
+}
+
+h1 {
+  font-size: 4em;
+  text-align: center;
+  margin-bottom: 0.5em;
+  margin-top: 2em;
+}
+
+h2 {
+  font-size: 2.2em;
+  font-weight:normal;
+  text-align: center;
+  margin-bottom: 48px;
+  margin-top: 0px;
+}
+
+.text_container {
+  margin: auto;
+  margin-top: 76px;
+  max-width: 600px;
+  line-height: 2em;
+  margin-bottom: 100px;
+}
+
+h3 {
+  font-size: 1.6em;
+  font-weight: bold;
+  /* text-align: center; */
+  margin-bottom: 0.5em;
+  margin-top: 4em;
+}
+
+
+h4 {
+  font-size: 1.2em;
+  /*font-weight: bold;*/
+  /* text-align: center; */
+  margin-bottom: 0.5em;
+  margin-top: 4em;
+}
+
+.code{
+  border: 0.5px solid lightgray;
+  background-color: GhostWhite;
+  border-radius: 3px;
+  padding-top: 4px;
+  padding-bottom: 4px;
+  padding-right: 12px;
+  padding-left: 12px;
+  color: gray;
+}
+
+.duration{
+  color: rgba(34,34,34,0.6);
+  font-style: italic;
+  /*margin-top: -20px;*/
+}
+
+ul {
+  line-height: 36px;
+}
+
+footer {
+  text-align: center;
+  font-size: 14px;
+  margin-top: 128px;
+  margin-bottom: 64px;
+  color: rgba(255, 255, 255, 0.4);
+}

Test_noise_reduction/noise_reduction-master/trim.py

+import librosa
+
+''' 
+01_counting.m4a
+02_wind_and_cars.m4a
+03_truck.m4a
+04_voices.m4a
+05_ambeint.m4a
+06_office.m4a
+'''
+
+sample_file = '06_office.m4a'
+sample_directory = '00_samples/'
+sample_path = sample_directory + sample_file
+trimmed_destination = 'samples_trimmed/'
+silenced_destination = 'samples_silence_reduced/'
+
+y, sr = librosa.load(sample_path)
+y_trimmed, index = librosa.effects.trim(y, top_db=12, frame_length=2)
+print(librosa.get_duration(y), librosa.get_duration(y_trimmed))
+
+destination = trimmed_destination + sample_file[:-4] + '.wav'
+librosa.output.write_wav(destination, y_trimmed, sr)