Experimental Auto Detect Batuk, Thread Record

.gitignore

@@ -1 +1,2 @@
 /.cache
+__pycache__

coughgui/pysources/experimental/auto_detect.py

@@ -0,0 +1,63 @@
+#pip install xgboost==0.90 scipy==0.22.1
+
+import alsaaudio as alsa
+import numpy as np
+from queue import Queue
+from datetime import datetime, timedelta
+from time import sleep
+import threading
+import os
+import wave
+import soundfile as sf
+
+from src.DSP import classify_cough
+import pickle
+
+model = pickle.load(open(os.path.join('./models', 'cough_classifier'), 'rb'))
+scaler = pickle.load(open(os.path.join('./models', 'cough_classification_scaler'), 'rb'))
+
+AudioLong = 1024
+to_checkAudio = np.zeros(3 * AudioLong, dtype='i2')
+
+aslsa_pcm = alsa.PCM(alsa.PCM_CAPTURE, alsa.PCM_NORMAL, channels=2, rate=44100,format=alsa.PCM_FORMAT_S16_LE, periodsize=AudioLong)
+
+data_queue = Queue()
+phrase_time = None
+phrase_timeout = 3
+
+def capture_audio():
+    phrase_time = datetime.utcnow()
+    total_indata = bytearray()
+    while True:
+        now = datetime.utcnow()
+        _, indata = aslsa_pcm.read()
+
+        if now - phrase_time > timedelta(seconds=phrase_timeout):
+            phrase_time = now
+            data_queue.put(bytes(total_indata))
+            total_indata = bytearray(indata)
+        else:
+            total_indata += indata
+
+capture_thread = threading.Thread(target=capture_audio)
+capture_thread.daemon = True
+capture_thread.start()
+
+while True:
+    try:
+        now = datetime.utcnow()
+        if not data_queue.empty():
+            audio_data = data_queue.get()
+            data_queue.queue.clear()
+
+            audio_np = np.frombuffer(audio_data, dtype=np.int16).astype(np.float32) / 32768.0
+            reshape_np = audio_np.reshape(-1, 2)
+
+            prob = classify_cough(reshape_np, 44100, model, scaler)
+            #print(f"has probability of cough: {prob}")
+            if prob > 0.75:
+                print(f"Batuk")            
+
+            sleep(0.25)
+    except KeyboardInterrupt:
+        break

coughgui/pysources/experimental/background_record.py

@@ -0,0 +1,63 @@
+import alsaaudio as alsa
+import numpy as np
+from queue import Queue
+from datetime import datetime, timedelta
+from time import sleep
+import threading
+import os
+import wave
+import soundfile as sf
+
+AudioLong = 1024
+to_checkAudio = np.zeros(3 * AudioLong, dtype='i2')
+
+aslsa_pcm = alsa.PCM(alsa.PCM_CAPTURE, alsa.PCM_NORMAL, channels=2, rate=44100,format=alsa.PCM_FORMAT_S16_LE, periodsize=AudioLong)
+
+data_queue = Queue()
+phrase_time = None
+phrase_timeout = 3
+
+def capture_audio():
+    phrase_time = datetime.utcnow()
+    total_indata = bytearray()
+    while True:
+        now = datetime.utcnow()
+        _, indata = aslsa_pcm.read()
+
+        if now - phrase_time > timedelta(seconds=phrase_timeout):
+            phrase_time = now
+            data_queue.put(bytes(total_indata))
+            total_indata = bytearray(indata)
+        else:
+            total_indata += indata
+
+capture_thread = threading.Thread(target=capture_audio)
+capture_thread.daemon = True
+capture_thread.start()
+
+i = 0
+
+while True:
+    try:
+        now = datetime.utcnow()
+        if not data_queue.empty():
+            audio_data = data_queue.get()
+            data_queue.queue.clear()
+
+            # Convert in-ram buffer to something the model can use directly without needing a temp file.
+            # Convert data from 16 bit wide integers to floating point with a width of 32 bits.
+            # Clamp the audio stream frequency to a PCM wavelength compatible default of 32768hz max.
+            audio_np = np.frombuffer(audio_data, dtype=np.int16).astype(np.float32) / 32768.0
+            reshape_np = audio_np.reshape(-1, 2)
+            #print(.shape)
+            sf.write(f"temp/coba_{i}.wav", reshape_np, 44100)
+            # with wave.open(f"temp/coba_{i}.wav", "wb") as out_wav:
+            #     out_wav.setparams((2, 2, 44100, 0, 'NONE', 'NONE'))
+            #     out_wav.writeframesraw(audio_data)
+
+            i += 1
+
+            # Infinite loops are bad for processors, must sleep.
+            sleep(0.25)
+    except KeyboardInterrupt:
+        break

coughgui/pysources/experimental/src/DSP.py

@@ -0,0 +1,79 @@
+import numpy as np
+import librosa
+from scipy import signal
+from scipy.io import wavfile
+from scipy.signal import butter,filtfilt
+from scipy.signal import cwt
+from scipy.signal import hilbert
+from scipy.signal import resample
+from scipy.signal import decimate
+from scipy.signal import spectrogram
+from scipy.signal.windows import get_window
+
+import os
+
+from .feature_class import features 
+
+
+def classify_cough(x, fs, model, scaler):
+    """Classify whether an inputted signal is a cough or not using filtering, feature extraction, and ML classification
+    Inputs: 
+        x: (float array) raw cough signal
+        fs: (int) sampling rate of raw signal
+        model: cough classification ML model loaded from file
+    Outputs:
+        result: (float) probability that a given file is a cough 
+    """
+    try: 
+        x,fs = preprocess_cough(x,fs)
+        data = (fs,x)
+        FREQ_CUTS = [(0,200),(300,425),(500,650),(950,1150),(1400,1800),(2300,2400),(2850,2950),(3800,3900)]
+        features_fct_list = ['EEPD','ZCR','RMSP','DF','spectral_features','SF_SSTD','SSL_SD','MFCC','CF','LGTH','PSD']
+        feature_values_vec = []
+        obj = features(FREQ_CUTS)
+        for feature in features_fct_list:
+            feature_values, feature_names = getattr(obj,feature)(data)
+            for value  in feature_values:
+                if isinstance(value,np.ndarray):
+                    feature_values_vec.append(value[0])
+                else:
+                    feature_values_vec.append(value)
+        feature_values_scaled = scaler.transform(np.array(feature_values_vec).reshape(1,-1))
+        result = model.predict_proba(feature_values_scaled)[:,1]
+        return result[0]
+    except:
+        "Feature extraction fails when the audio is completely silent"
+        return 0
+
+def preprocess_cough(x,fs, cutoff = 6000, normalize = True, filter_ = True, downsample = True):
+    """
+    Normalize, lowpass filter, and downsample cough samples in a given data folder 
+
+    Inputs: x*: (float array) time series cough signal
+    fs*: (int) sampling frequency of the cough signal in Hz
+    cutoff: (int) cutoff frequency of lowpass filter
+    normalize: (bool) normailzation on or off
+    filter: (bool) filtering on or off
+    downsample: (bool) downsampling on or off
+    *: mandatory input
+
+    Outputs: x: (float32 array) new preprocessed cough signal
+    fs: (int) new sampling frequency
+    """
+
+    fs_downsample = cutoff*2
+
+    #Preprocess Data
+    if len(x.shape)>1:
+        x = np.mean(x,axis=1)                          # Convert to mono
+    if normalize:
+        x = x/(np.max(np.abs(x))+1e-17)                # Norm to range between -1 to 1
+    if filter_:
+        b, a = butter(4, fs_downsample/fs, btype='lowpass') # 4th order butter lowpass filter
+        x = filtfilt(b, a, x)
+    if downsample:
+        x = signal.decimate(x, int(fs/fs_downsample)) # Downsample for anti-aliasing
+
+    fs_new = fs_downsample
+
+    return np.float32(x), fs_new

coughgui/pysources/experimental/src/convert_files.py

@@ -0,0 +1,24 @@
+import numpy as np
+import pandas as pd
+import os
+import subprocess
+from pathlib import Path
+
+
+def convert_files(folder):
+    """Convert files from .webm and .ogg to .wav
+    folder: path to coughvid database and metadata_compiled csv"""
+
+    df = pd.read_csv(folder + 'metadata_compiled.csv')
+    names_to_convert = df.uuid.to_numpy()
+    for counter, name in enumerate(names_to_convert):
+        if (counter%1000 == 0):
+            print("Finished {0}/{1}".format(counter,len(names_to_convert)))
+        if os.path.isfile(folder + name + '.webm'):
+            subprocess.call(["ffmpeg", "-i", folder+name+".webm", folder+name+".wav"])
+        elif os.path.isfile(folder + name + '.ogg'):
+            subprocess.call(["ffmpeg", "-i", folder+name+".ogg", folder+name+".wav"])
+        else:
+            print("Error: No file name {0}".format(name))
+
+