{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"markdown","source":"## Import the libraries\n","metadata":{"execution":{"iopub.status.busy":"2022-02-11T15:15:42.09471Z","iopub.execute_input":"2022-02-11T15:15:42.095103Z","iopub.status.idle":"2022-02-11T15:15:42.099724Z","shell.execute_reply.started":"2022-02-11T15:15:42.095016Z","shell.execute_reply":"2022-02-11T15:15:42.098645Z"}}},{"cell_type":"code","source":"!pip install pyunpack\n!pip install patool\n!pip install py7zr\n!pip install sounddevice\n!pip install noisereduce\n!pip install librosa\n! pip install python_speech_features\n! pip install tensorflow==2.4\n! pip install malaya_speech\n! pip install webrtcvad","metadata":{"execution":{"iopub.status.busy":"2022-06-24T23:54:12.653899Z","iopub.execute_input":"2022-06-24T23:54:12.654265Z","iopub.status.idle":"2022-06-24T23:56:24.402429Z","shell.execute_reply.started":"2022-06-24T23:54:12.654186Z","shell.execute_reply":"2022-06-24T23:56:24.401513Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"pip install py7zr","metadata":{"execution":{"iopub.status.busy":"2022-06-24T23:56:24.405681Z","iopub.execute_input":"2022-06-24T23:56:24.405951Z","iopub.status.idle":"2022-06-24T23:56:30.973268Z","shell.execute_reply.started":"2022-06-24T23:56:24.405923Z","shell.execute_reply":"2022-06-24T23:56:30.972350Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#pip update huggingface_hub\n! pip install --upgrade transformers","metadata":{"execution":{"iopub.status.busy":"2022-06-24T23:56:30.974982Z","iopub.execute_input":"2022-06-24T23:56:30.975236Z","iopub.status.idle":"2022-06-24T23:56:44.217869Z","shell.execute_reply.started":"2022-06-24T23:56:30.975209Z","shell.execute_reply":"2022-06-24T23:56:44.216912Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **importing libraries**","metadata":{}},{"cell_type":"code","source":"pip install git+https://github.com/huggingface/transformers","metadata":{"execution":{"iopub.status.busy":"2022-06-24T23:56:44.221074Z","iopub.execute_input":"2022-06-24T23:56:44.221342Z","iopub.status.idle":"2022-06-24T23:57:26.049802Z","shell.execute_reply.started":"2022-06-24T23:56:44.221314Z","shell.execute_reply":"2022-06-24T23:57:26.048746Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"import numpy as np                        # linear algebra library\nimport pandas as pd                       # data frames processing\nimport matplotlib.pyplot as plt          # visualization library\nimport seaborn as sn                      # visualization library\n\n\n# audio processing library\nimport librosa                          \nimport IPython.display as ipd            \nfrom scipy.io import wavfile\nimport noisereduce as nr\nfrom malaya_speech import Pipeline\nimport malaya_speech\nfrom python_speech_features import mfcc\nfrom sklearn.preprocessing import LabelEncoder\n\n\n# unpacking the dataset\nfrom py7zr import unpack_7zarchive    \n\n#operating system libraries\nimport shutil\nimport warnings\nwarnings.filterwarnings(\"ignore\")\nimport tensorflow \nimport os","metadata":{"execution":{"iopub.status.busy":"2022-06-24T23:57:26.051247Z","iopub.execute_input":"2022-06-24T23:57:26.051512Z","iopub.status.idle":"2022-06-24T23:57:31.052569Z","shell.execute_reply.started":"2022-06-24T23:57:26.051483Z","shell.execute_reply":"2022-06-24T23:57:31.051700Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# unpacking the dataset\nshutil.register_unpack_format('7zip', ['.7z'], unpack_7zarchive)\nshutil.unpack_archive('/kaggle/input/tensorflow-speech-recognition-challenge/train.7z', '/kaggle/working/tensorflow-speech-recognition-challenge/train/')","metadata":{"execution":{"iopub.status.busy":"2022-06-24T23:57:31.053957Z","iopub.execute_input":"2022-06-24T23:57:31.054281Z","iopub.status.idle":"2022-06-25T00:06:48.920040Z","shell.execute_reply.started":"2022-06-24T23:57:31.054247Z","shell.execute_reply":"2022-06-25T00:06:48.919162Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"# <center> Implementing the Speech Recognition Model in Python\n**Dataset used for our Speech Recognition Project**\n    \nIt is a set of 10 numbers each is repeated 2000 times with different accents and different back ground conditions. TensorFlow recently released the Speech Commands Datasets. It includes 65,000 one-second long utterances of 30 short words, by thousands of different people. We’ll build a speech recognition system that understands simple spoken commands. <br>    \n    \n__You can download the dataset from__ [here](https://www.kaggle.com/c/tensorflow-speech-recognition-challenge).\n","metadata":{}},{"cell_type":"markdown","source":"**Data Exploration and Visualization**\n\nData Exploration and Visualization helps us to understand the data as well as pre-processing steps in a better way. \n\n","metadata":{}},{"cell_type":"code","source":"train_audio_path = '/kaggle/working/tensorflow-speech-recognition-challenge/train/train/audio/' #path of the training data","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:09:20.372380Z","iopub.execute_input":"2022-06-25T00:09:20.372785Z","iopub.status.idle":"2022-06-25T00:09:20.377086Z","shell.execute_reply.started":"2022-06-25T00:09:20.372748Z","shell.execute_reply":"2022-06-25T00:09:20.375928Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Accessing a sample file in data**","metadata":{}},{"cell_type":"code","source":"samples, sample_rate = librosa.load(train_audio_path+'on/5a3712c9_nohash_1.wav', sr = 16000)  # loading a sample to be explored carefully\nfig = plt.figure(figsize=(14, 8))\nax1 = fig.add_subplot(211)\nax1.set_title('Raw wave of ' + '../input/train/audio/on/0a7c2a8d_nohash_0.wav')\nax1.set_xlabel('time')\nax1.set_ylabel('Amplitude')\nax1.plot(np.linspace(0, sample_rate/len(samples), sample_rate), samples)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:09:22.017406Z","iopub.execute_input":"2022-06-25T00:09:22.017872Z","iopub.status.idle":"2022-06-25T00:09:22.202225Z","shell.execute_reply.started":"2022-06-25T00:09:22.017834Z","shell.execute_reply":"2022-06-25T00:09:22.201409Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"fs=16000\nipd.Audio(samples, rate=fs)  #listen to audio file before noise reduction\nprint(\"Sample Rate:\",fs)\nsr=fs","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:15.157752Z","iopub.execute_input":"2022-06-25T00:10:15.158087Z","iopub.status.idle":"2022-06-25T00:10:15.166546Z","shell.execute_reply.started":"2022-06-25T00:10:15.158058Z","shell.execute_reply":"2022-06-25T00:10:15.165537Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Noise Reduction\ntime = np.linspace(0, len(samples - 1) / fs, len(samples - 1))\nreduced_noise1 = nr.reduce_noise(y=samples, sr=fs,stationary=True)\nplt.plot(time, reduced_noise1)  # plot in seconds\nplt.xlabel(\"Time [seconds]\")\nplt.ylabel(\"Voice amplitude\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:16.061777Z","iopub.execute_input":"2022-06-25T00:10:16.062090Z","iopub.status.idle":"2022-06-25T00:10:16.769569Z","shell.execute_reply.started":"2022-06-25T00:10:16.062061Z","shell.execute_reply":"2022-06-25T00:10:16.768692Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ipd.Audio(reduced_noise1, rate=sample_rate)  #listen to audio file after noise reduction","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:16.770995Z","iopub.execute_input":"2022-06-25T00:10:16.771499Z","iopub.status.idle":"2022-06-25T00:10:16.781272Z","shell.execute_reply.started":"2022-06-25T00:10:16.771458Z","shell.execute_reply":"2022-06-25T00:10:16.780347Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"#Silence Removal\nvad = malaya_speech.vad.webrtc()\ny=reduced_noise1\ny_= malaya_speech.resample(y, sr, 16000)\ny_ = malaya_speech.astype.float_to_int(y_)\nframes = malaya_speech.generator.frames(y, 30, sr)\nframes_ = list(malaya_speech.generator.frames(y_, 30, 16000, append_ending_trail = False))\nframes_webrtc = [(frames[no], vad(frame)) for no, frame in enumerate(frames_)]\ny_ = malaya_speech.combine.without_silent(frames_webrtc)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:17.402023Z","iopub.execute_input":"2022-06-25T00:10:17.402373Z","iopub.status.idle":"2022-06-25T00:10:17.419266Z","shell.execute_reply.started":"2022-06-25T00:10:17.402339Z","shell.execute_reply":"2022-06-25T00:10:17.418449Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"ipd.Audio(y_, rate = sr )     #listen to audio file after noise reduction","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:17.708750Z","iopub.execute_input":"2022-06-25T00:10:17.709055Z","iopub.status.idle":"2022-06-25T00:10:17.716685Z","shell.execute_reply.started":"2022-06-25T00:10:17.709027Z","shell.execute_reply":"2022-06-25T00:10:17.715494Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# padding signal with zeros\nzero = np.zeros((1*sr-y_.shape[0]))\nsignal = np.concatenate((y_,zero))\nsignal.shape\ntime = np.linspace(0, len(signal - 1) / fs, len(signal - 1))","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:17.944854Z","iopub.execute_input":"2022-06-25T00:10:17.945144Z","iopub.status.idle":"2022-06-25T00:10:17.950377Z","shell.execute_reply.started":"2022-06-25T00:10:17.945117Z","shell.execute_reply":"2022-06-25T00:10:17.949304Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"**Visualization of Audio signal in time series domain**\n\nNow, we’ll visualize the audio signal in the time series domain:","metadata":{}},{"cell_type":"code","source":"plt.plot(time,signal)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:18.968172Z","iopub.execute_input":"2022-06-25T00:10:18.968559Z","iopub.status.idle":"2022-06-25T00:10:19.110127Z","shell.execute_reply.started":"2022-06-25T00:10:18.968524Z","shell.execute_reply":"2022-06-25T00:10:19.109294Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"labels=os.listdir(train_audio_path)     #Extracting labels to determine classes","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:19.349750Z","iopub.execute_input":"2022-06-25T00:10:19.350069Z","iopub.status.idle":"2022-06-25T00:10:19.354901Z","shell.execute_reply.started":"2022-06-25T00:10:19.350040Z","shell.execute_reply":"2022-06-25T00:10:19.353789Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **Distribution of the Data set**","metadata":{}},{"cell_type":"code","source":"#find count of each label and plot bar graph\nno_of_recordings=[]\nfor label in labels:\n    waves = [f for f in os.listdir(train_audio_path + '/'+ label) if f.endswith('.wav')]\n    no_of_recordings.append(len(waves))\n    \n#plot\nplt.figure(figsize=(30,5))\nindex = np.arange(len(labels))\nplt.bar(index, no_of_recordings)\nplt.xlabel('Commands', fontsize=12)\nplt.ylabel('No of recordings', fontsize=12)\nplt.xticks(index, labels, fontsize=15, rotation=60)\nplt.title('No. of recordings for each command')\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:20.016989Z","iopub.execute_input":"2022-06-25T00:10:20.017402Z","iopub.status.idle":"2022-06-25T00:10:20.351263Z","shell.execute_reply.started":"2022-06-25T00:10:20.017359Z","shell.execute_reply":"2022-06-25T00:10:20.350279Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Words used as Classes for data set","metadata":{}},{"cell_type":"code","source":"labels=[\"zero\",\"one\",\"two\",\"three\",\"four\",\"five\",\"six\",\"seven\",\"eight\",\"nine\"]","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:20.559792Z","iopub.execute_input":"2022-06-25T00:10:20.560093Z","iopub.status.idle":"2022-06-25T00:10:20.564573Z","shell.execute_reply.started":"2022-06-25T00:10:20.560065Z","shell.execute_reply":"2022-06-25T00:10:20.563420Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **Preprocessing the audio waves**\n\nlet us read the audio waves and use the below-preprocessing steps :\n\n* Noise Reduction\n* Silence Removal\n* Extracting MFCCs\n","metadata":{}},{"cell_type":"code","source":"sr=16000    # sample rate\nvad = malaya_speech.vad.webrtc()\nall_wave = []     #intitialize array to stack wave files of the whole data set in it \nall_label = []    #intitialize array to stack label of wave files of the whole data set in it \nfor label in labels:\n    print(label)\n    waves = [f for f in os.listdir(train_audio_path + '/'+ label) if f.endswith('.wav')] # access on each file\n    for wav in waves:\n        samples, sample_rate = librosa.load(train_audio_path + '/' + label + '/' + wav, sr = 16000)\n        samples = nr.reduce_noise(y=samples, sr=sr,stationary=True)  #noise reduction\n        y_= malaya_speech.resample(samples, sr, 16000)               # silence removal\n        y_ = malaya_speech.astype.float_to_int(y_)\n        frames = malaya_speech.generator.frames(samples, 30, sr)\n        frames_ = list(malaya_speech.generator.frames(y_, 30, 16000, append_ending_trail = False))\n        frames_webrtc = [(frames[no], vad(frame)) for no, frame in enumerate(frames_)]\n        y_ = malaya_speech.combine.without_silent(frames_webrtc)\n        zero = np.zeros(((1*sr+4000)-y_.shape[0]))                 \n        signal = np.concatenate((y_,zero))     # concatenation with zeros to adust length of the vector\n        all_wave.append(signal)     #append waves one by one \n        all_label.append(label)     #append corresponding label one by one","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:10:21.777403Z","iopub.execute_input":"2022-06-25T00:10:21.777797Z","iopub.status.idle":"2022-06-25T00:21:08.650952Z","shell.execute_reply.started":"2022-06-25T00:10:21.777761Z","shell.execute_reply":"2022-06-25T00:21:08.650080Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"shape of waves array\",np.array(all_wave).shape)\nprint(\"shape of labels array\",np.array(all_label).shape)\n\n#Inspecting random sample\ntime = np.linspace(0, len(signal - 1) / fs, len(signal - 1))\nplt.plot(time,np.array(all_wave)[2000,:])\nprint(np.array(all_label)[2000])\nipd.Audio(np.array(all_wave)[2000,:], rate = sr )","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:21:08.652360Z","iopub.execute_input":"2022-06-25T00:21:08.652751Z","iopub.status.idle":"2022-06-25T00:21:19.813146Z","shell.execute_reply.started":"2022-06-25T00:21:08.652711Z","shell.execute_reply":"2022-06-25T00:21:19.812315Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## Extracting MFCCs","metadata":{}},{"cell_type":"code","source":"all_mfcc=[]    #intitialize array to stack MFCCs of the whole data set in it\nfor wave in all_wave:\n    i=0\n    mfcc_feat = mfcc(wave , fs, winlen=256/fs, winstep=256/(2*fs), numcep=13, nfilt=26, nfft=256,\n                 lowfreq=0, highfreq=fs/2, preemph=0.97, ceplifter=22, appendEnergy=True, winfunc=np.hamming)\n    mfcc_feat= np.transpose(mfcc_feat)\n    all_mfcc.append(mfcc_feat)","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:21:19.815069Z","iopub.execute_input":"2022-06-25T00:21:19.815431Z","iopub.status.idle":"2022-06-25T00:22:28.513321Z","shell.execute_reply.started":"2022-06-25T00:21:19.815393Z","shell.execute_reply":"2022-06-25T00:22:28.512202Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"shape of MFCCs\",np.array(all_mfcc).shape)\nprint(\"shape of Corresponding lables\",np.array(all_label).shape)\n\n# dimensions of the data\nd1=np.array(all_mfcc).shape[1]\nd2=np.array(all_mfcc).shape[2]\nd=d1*d2","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:22:28.514714Z","iopub.execute_input":"2022-06-25T00:22:28.515325Z","iopub.status.idle":"2022-06-25T00:22:29.130980Z","shell.execute_reply.started":"2022-06-25T00:22:28.515286Z","shell.execute_reply":"2022-06-25T00:22:29.129842Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"op_mfcc=np.array(all_mfcc)     # transform list to array to be fed to the model\nop_mfcc=op_mfcc.reshape(np.array(all_mfcc).shape[0],-1)   # adjust shape of the samples\nop_mfcc.shape","metadata":{"execution":{"iopub.status.busy":"2022-06-25T00:22:29.132672Z","iopub.execute_input":"2022-06-25T00:22:29.133000Z","iopub.status.idle":"2022-06-25T00:22:29.499378Z","shell.execute_reply.started":"2022-06-25T00:22:29.132963Z","shell.execute_reply":"2022-06-25T00:22:29.498481Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## One hot encoding","metadata":{}},{"cell_type":"code","source":"#all_label = all_label.tolist()\nle = LabelEncoder()\ny=le.fit_transform(all_label)\nclasses= list(le.classes_)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:50:36.474324Z","iopub.execute_input":"2022-05-17T17:50:36.474694Z","iopub.status.idle":"2022-05-17T17:50:36.490416Z","shell.execute_reply.started":"2022-05-17T17:50:36.474647Z","shell.execute_reply":"2022-05-17T17:50:36.489597Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"## **Model based on ANN** ","metadata":{}},{"cell_type":"code","source":"! pip install --upgrade tensorflow\n! pip install --upgrade tensorflow-gpu\n! pip install keras==2.3.1","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:50:36.49465Z","iopub.execute_input":"2022-05-17T17:50:36.494902Z","iopub.status.idle":"2022-05-17T17:52:48.165204Z","shell.execute_reply.started":"2022-05-17T17:50:36.494877Z","shell.execute_reply":"2022-05-17T17:52:48.164174Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Importing model libraries from keras\nfrom keras.optimizers import SGD\nfrom keras.constraints import maxnorm\nfrom tensorflow.keras import Sequential\nfrom tensorflow.keras.layers import Flatten, Dense,Dropout\nfrom keras.callbacks import EarlyStopping, ModelCheckpoint\nfrom sklearn.model_selection import train_test_split\n","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:48.169422Z","iopub.execute_input":"2022-05-17T17:52:48.169762Z","iopub.status.idle":"2022-05-17T17:52:48.228057Z","shell.execute_reply.started":"2022-05-17T17:52:48.169723Z","shell.execute_reply":"2022-05-17T17:52:48.227092Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"y=tensorflow.keras.utils.to_categorical(y, num_classes=len(labels), dtype='float32')  # one hot encoded varibales to categeorical values\ny.shape","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:48.229653Z","iopub.execute_input":"2022-05-17T17:52:48.230265Z","iopub.status.idle":"2022-05-17T17:52:48.239535Z","shell.execute_reply.started":"2022-05-17T17:52:48.230221Z","shell.execute_reply":"2022-05-17T17:52:48.238525Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"x_tr, x_val, y_tr, y_val= train_test_split(op_mfcc,np.array(y),stratify=y,test_size = 0.2,random_state=777,shuffle=True)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:48.241037Z","iopub.execute_input":"2022-05-17T17:52:48.241414Z","iopub.status.idle":"2022-05-17T17:52:48.989901Z","shell.execute_reply.started":"2022-05-17T17:52:48.241376Z","shell.execute_reply":"2022-05-17T17:52:48.988988Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"# Inspect shapes of each set\nprint(\"shape of training samples\",x_tr.shape)\nprint(\"shape of training labels\",y_tr.shape)\nprint(\"shape of test samples\",x_val.shape)\nprint(\"shape of test labels\",y_val.shape)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:48.991095Z","iopub.execute_input":"2022-05-17T17:52:48.99144Z","iopub.status.idle":"2022-05-17T17:52:48.998308Z","shell.execute_reply.started":"2022-05-17T17:52:48.991403Z","shell.execute_reply":"2022-05-17T17:52:48.997385Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### **Model Architecture**","metadata":{}},{"cell_type":"code","source":"#Model Architecture\nmodel = Sequential()\nmodel.add(Dense(100, activation='relu', input_shape=(d,), kernel_constraint=maxnorm(3)))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(80, activation='relu', kernel_constraint=maxnorm(3)))\nmodel.add(Dropout(0.5))\nmodel.add(Dense(len(classes), activation='softmax' , kernel_constraint=maxnorm(3)))","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:48.999662Z","iopub.execute_input":"2022-05-17T17:52:49.000343Z","iopub.status.idle":"2022-05-17T17:52:56.328984Z","shell.execute_reply.started":"2022-05-17T17:52:49.000283Z","shell.execute_reply":"2022-05-17T17:52:56.328048Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"tensorflow.keras.utils.plot_model(model, 'model.png',show_shapes=True)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:56.33038Z","iopub.execute_input":"2022-05-17T17:52:56.330772Z","iopub.status.idle":"2022-05-17T17:52:57.237802Z","shell.execute_reply.started":"2022-05-17T17:52:56.330732Z","shell.execute_reply":"2022-05-17T17:52:57.236782Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"model.compile(loss='categorical_crossentropy',optimizer='adamax',metrics=['accuracy'])\n","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:57.239686Z","iopub.execute_input":"2022-05-17T17:52:57.240088Z","iopub.status.idle":"2022-05-17T17:52:57.261455Z","shell.execute_reply.started":"2022-05-17T17:52:57.240044Z","shell.execute_reply":"2022-05-17T17:52:57.260467Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"es = EarlyStopping(monitor='val_loss', mode='min', verbose=1, patience=10, min_delta=0.0001) \nmc = ModelCheckpoint('best_model.hdf5', monitor='val_acc', verbose=1, save_best_only=True, mode='max')","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:57.262891Z","iopub.execute_input":"2022-05-17T17:52:57.263622Z","iopub.status.idle":"2022-05-17T17:52:57.269088Z","shell.execute_reply.started":"2022-05-17T17:52:57.263579Z","shell.execute_reply":"2022-05-17T17:52:57.268205Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history=model.fit(x_tr, y_tr,validation_data=(x_val,y_val), epochs=300, batch_size=65)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:52:57.283514Z","iopub.execute_input":"2022-05-17T17:52:57.283978Z","iopub.status.idle":"2022-05-17T17:58:04.760272Z","shell.execute_reply.started":"2022-05-17T17:52:57.283927Z","shell.execute_reply":"2022-05-17T17:58:04.759107Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Model Evaluation","metadata":{}},{"cell_type":"code","source":"train_score = model.evaluate(x_tr, y_tr, batch_size=12)\nprint(train_score)\n\nprint('----------------Training Complete-----------------')\n\ntest_score = model.evaluate(x_val, y_val, batch_size = 12)\nprint(test_score)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:04.773583Z","iopub.execute_input":"2022-05-17T17:58:04.774293Z","iopub.status.idle":"2022-05-17T17:58:08.591167Z","shell.execute_reply.started":"2022-05-17T17:58:04.774229Z","shell.execute_reply":"2022-05-17T17:58:08.590245Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"history.history.keys()","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:08.592547Z","iopub.execute_input":"2022-05-17T17:58:08.592948Z","iopub.status.idle":"2022-05-17T17:58:08.599188Z","shell.execute_reply.started":"2022-05-17T17:58:08.592909Z","shell.execute_reply":"2022-05-17T17:58:08.598164Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['loss'], label='train')  # losses learning curve of training set.\nplt.plot(history.history['val_loss'], label='test') # losses learning curve of validation set.\nplt.legend()\nplt.title(\"losses learning curves\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:08.600726Z","iopub.execute_input":"2022-05-17T17:58:08.601194Z","iopub.status.idle":"2022-05-17T17:58:08.754097Z","shell.execute_reply.started":"2022-05-17T17:58:08.601154Z","shell.execute_reply":"2022-05-17T17:58:08.753107Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"plt.plot(history.history['accuracy'])      # Accuracy learning curve of training set.\nplt.plot(history.history['val_accuracy'])  # Accuracy learning curve of validation set.\nplt.title('model accuracy')\nplt.ylabel('accuracy')\nplt.xlabel('epoch')\nplt.legend(['train', 'val'], loc='upper left')\nplt.title(\"Accuracy learning curves\")\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:08.755648Z","iopub.execute_input":"2022-05-17T17:58:08.756046Z","iopub.status.idle":"2022-05-17T17:58:08.926267Z","shell.execute_reply.started":"2022-05-17T17:58:08.756001Z","shell.execute_reply":"2022-05-17T17:58:08.92523Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### **confusion matrix**","metadata":{}},{"cell_type":"code","source":"y_predict=model.predict(x_val)\nconf_mat=tensorflow.math.confusion_matrix(np.argmax(y_val,axis=1) , np.argmax(y_predict,axis=1))","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:08.927664Z","iopub.execute_input":"2022-05-17T17:58:08.928181Z","iopub.status.idle":"2022-05-17T17:58:09.614379Z","shell.execute_reply.started":"2022-05-17T17:58:08.928137Z","shell.execute_reply":"2022-05-17T17:58:09.613324Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"df_cm = pd.DataFrame(np.array(conf_mat), index = [i for i in classes],\n                  columns = [i for i in classes])\nplt.figure(figsize = (13,7))\nax = sn.heatmap(df_cm, annot=True)\nplt.title(\"Confusion Matrix\", fontsize=20)\nplt.ylabel(\"True Class\"     , fontsize=20)\nplt.xlabel(\"Predicted Class\", fontsize=20)\nplt.show()","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:09.615986Z","iopub.execute_input":"2022-05-17T17:58:09.616376Z","iopub.status.idle":"2022-05-17T17:58:10.403031Z","shell.execute_reply.started":"2022-05-17T17:58:09.616333Z","shell.execute_reply":"2022-05-17T17:58:10.402127Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"### Test random samples to check model performance","metadata":{}},{"cell_type":"code","source":"model.predict_classes(X_test)\nfrom sklearn.metrics import classification_report\nprint(classification_report(y_test, to_categorical(predictions)))","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:10.406243Z","iopub.execute_input":"2022-05-17T17:58:10.409286Z","iopub.status.idle":"2022-05-17T17:58:10.452113Z","shell.execute_reply.started":"2022-05-17T17:58:10.409241Z","shell.execute_reply":"2022-05-17T17:58:10.44978Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"x_val[1].shape\nmodel.predict(x_val[1].reshape((1,d)))","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:10.454817Z","iopub.status.idle":"2022-05-17T17:58:10.457141Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Define the function that predicts text for the given audio:","metadata":{}},{"cell_type":"code","source":"def predict(audio):\n    print(samples.shape)\n    prob=model.predict(audio)\n    index=np.argmax(prob[0])\n    return classes[index]","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:10.460805Z","iopub.status.idle":"2022-05-17T17:58:10.463065Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":"Prediction time! Make predictions on the validation data:","metadata":{}},{"cell_type":"code","source":"import random\nindex=random.randint(0,len(x_val)-1)\nprint(index)\nsamples=x_val[index]\nprint(\"Audio:\",classes[np.argmax(y_val[index])])\n#ipd.Audio(np.array(all_wave)[index,:], rate=16000)","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:10.466646Z","iopub.status.idle":"2022-05-17T17:58:10.468915Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"print(\"Text:\",predict(samples.reshape(1,d)))","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:10.470153Z","iopub.status.idle":"2022-05-17T17:58:10.474592Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"from keras.models import load_model\nmodel.save(\"SR_MODEL.h5\")    #loading model to be tested locally.","metadata":{"execution":{"iopub.status.busy":"2022-05-17T17:58:10.475863Z","iopub.status.idle":"2022-05-17T17:58:10.47667Z"},"trusted":true},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]},{"cell_type":"code","source":"","metadata":{},"execution_count":null,"outputs":[]}]}