Update app.py

app.py

@@ -2,7 +2,6 @@ import gradio as gr
 import tensorflow as tf
 import numpy as np
 import nltk
-import soundfile as sf
 from nltk.corpus import cmudict
 from scipy.io.wavfile import write
 
@@ -17,8 +16,6 @@ model = tf.keras.models.load_model('audio_model.h5')
 def preprocess_text(text):
     """
     Process the input text to prepare it for the model.
-    This could include tokenization, phoneme extraction, etc.
-    The model expects input of shape (batch_size, sequence_length, 13).
     """
     d = cmudict.dict()
     words = text.lower().split()
@@ -28,19 +25,19 @@ def preprocess_text(text):
         if word in d:
             phonemes.append(d[word][0])
         else:
-            # If word not found in cmudict, use a placeholder or skip
+            # Use a placeholder for words not found in cmudict
             phonemes.append(['UNKNOWN'])
     
     # Flatten the list of phonemes
     flattened_phonemes = [p for sublist in phonemes for p in sublist]
     
-    # Create dummy 13-feature vectors for each phoneme (you need to implement your own feature extraction)
+    # Create dummy feature vectors (this should be replaced with actual feature extraction)
     num_features = 13
     sequence_length = len(flattened_phonemes)
-    input_data = np.random.rand(sequence_length, num_features)  # Placeholder, replace with actual feature extraction
+    input_data = np.random.rand(sequence_length, num_features)  # Placeholder
     
     # Add batch dimension
-    input_data = np.expand_dims(input_data, axis=0)  # Shape (1, sequence_length, 13)
+    input_data = np.expand_dims(input_data, axis=0)
     
     return input_data
 
@@ -48,17 +45,20 @@ def preprocess_text(text):
 def convert_to_audio(model_output, filename="output.wav", sample_rate=22050):
     """
     Convert the model output into a .wav file.
-    Model output is expected to be a numpy array.
     """
-    # Normalize the audio output (optional, based on your model)
+    # Check if model_output is empty
+    if model_output is None or len(model_output) == 0:
+        raise ValueError("Model output is empty.")
+
+    # Normalize the audio output
     normalized_output = np.interp(model_output, (model_output.min(), model_output.max()), (-1, 1))
 
-    # Write the audio data to a file (assuming model_output is a waveform)
+    # Write the audio data to a file
     write(filename, sample_rate, normalized_output)
 
     return filename
 
-# Define function to generate sound effect
+# Generate sound effect
 def generate_sfx(text):
     """
     Takes input text, preprocesses it, runs it through the model,
@@ -69,8 +69,12 @@ def generate_sfx(text):
     # Generate prediction
     prediction = model.predict(input_data)
 
+    # Ensure prediction shape is correct
+    if prediction.ndim == 2 and prediction.shape[1] > 1:
+        prediction = prediction.flatten()  # Flatten if necessary
+
     # Convert the prediction to an audio file
-    audio_file = convert_to_audio(prediction.flatten(), filename="output.wav")
+    audio_file = convert_to_audio(prediction, filename="output.wav")
     
     return audio_file
 
@@ -78,7 +82,7 @@ def generate_sfx(text):
 interface = gr.Interface(
     fn=generate_sfx,
     inputs=gr.Textbox(label="Enter a Word", placeholder="Write a Word To Convert it into SFX Sound"),
-    outputs=gr.Audio(label="Generated SFX", type="filepath"),  # Changed here
+    outputs=gr.Audio(label="Generated SFX", type="filepath"),
     live=False,
     title="SFX Generator from Text",
     description="Enter a word or sentence, and the model will generate an SFX sound.",