I already built a keras model that can will detect the conditions by their audio using microphone, and I want to try the model by using it with gradio interface but unsuccessful.
this is my model architecture:
import tensorflow as tf
from keras.models import Sequential
from keras.layers import Conv1D, MaxPooling1D, Flatten, Dense, Dropout
from keras.utils import to_categorical
input_shape = (spectrograms.shape[1], 1)
model = Sequential()
model.add(Conv1D(32, 3, padding='valid', strides=1, input_shape=input_shape, activation='relu'))
model.add(MaxPooling1D(pool_size=4, strides=3))
model.add(Conv1D(32, 1, padding='valid', strides=1, activation='relu'))
model.add(MaxPooling1D(pool_size=1, strides=3))
model.add(Flatten())
model.add(Dense(512, activation='relu'))
model.add(Dense(512, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(2, activation='sigmoid')) # Assuming 2 classes for binary classification encoded
#model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy',
tf.keras.metrics.Precision(),
tf.keras.metrics.Recall(),
tf.keras.metrics.F1Score(),
tf.keras.metrics.TrueNegatives(),
tf.keras.metrics.TruePositives(),
tf.keras.metrics.FalseNegatives(),
tf.keras.metrics.FalsePositives()])
and from this point
import os
import numpy as np
import librosa
import librosa.display
import matplotlib.pyplot as plt
from keras.utils import to_categorical
# Load audio data and create spectrograms
def predict_conditions(audio):
audio_data, _ = librosa.load(audio, sr=sample_rate, duration=duration)
sample_rate=22050,
duration=4,
hop_length=512,
n_mels=128
plt.figure(figsize=(10, 4))
mel_spectrogram = librosa.feature.melspectrogram(y=audio_data, sr=sample_rate, hop_length=hop_length, n_mels=n_mels)
librosa.display.specshow(librosa.power_to_db(mel_spectrogram, ref=np.max), y_axis='mel', x_axis='time')
plt.colorbar(format='%+2.0f dB')
plt.title('Mel Spectrogram')
# Extract the file name without extension
file_name = os.path.splitext(os.path.basename(audio_path))[0]
# Create the output directory if it doesn't exist
if not os.path.exists(output_dir):
os.makedirs(output_dir)
# Specify the path to save the spectrogram image
image_path = os.path.join(output_dir, f'{file_name}_spectrogram.png')
# Save the spectrogram as an image
plt.savefig(image_path, bbox_inches='tight', pad_inches=0)
plt.close()
output_dir = 'D:/dataset/something/spectrogram'
return mel_spectrogram, image_path
# Define the directory where your spectrogram images are stored
output_dir = 'D:/dataset/something/spectrogram'
# Create empty lists to store spectrograms
spectrograms = []
# Function to load spectrogram from image path
def load_spectrogram(image_path, sample_rate=22050, duration=4, hop_length=512, n_mels=128):
# Load the image
image = plt.imread(image_path)
# Convert the image to grayscale if needed
if len(image.shape) == 3:
image = image.mean(axis=-1)
# Perform inverse of the preprocessing steps in create_spectrogram
mel_spectrogram = librosa.feature.inverse.mel_to_stft(image)
audio_data = librosa.feature.inverse.mel_to_audio(mel_spectrogram, sr=sample_rate, hop_length=hop_length)
return audio_data
# Load spectrograms and labels
for root, dirs, files in os.walk(spectrogram_dir):
for file in files:
if file.endswith("_spectrogram.png"):
image_path = os.path.join(root, file)
spectrogram = load_spectrogram(image_path) # Load the spectrogram image
spectrograms.append(spectrogram)
# Prepare the spectrogram
spectrograms = np.array(spectrograms)
until this point above is my pre-processing steps. And because of that, I thought if I want to test it with gradio, the data needs to be undergoing the similar process as the training. The only difference is that, the training will load the data from directory but for the gradio will take the audio directly from mic. It continues with code below to display the gradio interface. Anyone know what I missed?
def convert_predictions_to_binary(predictions):
average_prediction = np.mean(predictions)
threshold = 0.5
return "okay" if average_prediction > threshold else "not okay"
predictions = model.predict(spectrograms)
# Make binary predictions for each spectrogram and print them to the console.
for prediction in predictions:
binary_prediction = convert_predictions_to_binary(prediction)
print(binary_prediction)
test = gr.Interface(fn=predict_conditions, inputs="microphone", outputs = gr.Label(num_top_classes=2),
title=title, description=description)
if __name__ == "__main__":
test.launch()
Sorry for inconvenience occurs, I’m just a newbie here. I greatly appreciate the help