Update app.py

app.py

@@ -4,58 +4,49 @@ import torch
 from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
 from datasets import load_dataset
 import soundfile as sf
-import io
 
-# Load models and pronunciation dictionary
+# Step 3: Load the models and the pronunciation dictionary
 processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
 model = SpeechT5ForTextToSpeech.from_pretrained("microsoft/speecht5_tts")
 vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
 
+# Load pronunciation dictionary from JSON file
 with open("pronunciation_dict.json", "r") as f:
     pronunciation_dict = json.load(f)
 
-# Preprocess the text
+# Function to preprocess the input text
 def preprocess_text(text):
     for term, phonetic in pronunciation_dict.items():
         text = text.replace(term, phonetic)
     return text
 
-# Text-to-Speech function
+# Step 4: Define the TTS function
 def text_to_speech(input_text):
+    # Preprocess the text
     processed_text = preprocess_text(input_text)
+
+    # Convert the processed text to model inputs
     inputs = processor(text=processed_text, return_tensors="pt")
-    
+
+    # Load xvector embeddings from dataset for speaker voice characteristics
     embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
     speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
-    
+
+    # Generate speech using the model and vocoder
     speech = model.generate_speech(inputs["input_ids"], speaker_embeddings, vocoder=vocoder)
-    
-    # Prepare audio as BytesIO
-    audio_buffer = io.BytesIO()
-    sf.write(audio_buffer, speech.numpy(), samplerate=16000, format='WAV')
-    audio_buffer.seek(0)
-    
-    return audio_buffer
-
-# Define examples
-examples = [
-    "We are using API for authentication.",
-    "CUDA and TensorFlow work together for deep learning models.",
-    "The database uses NoSQL and supports JSON for data storage.",
-    "Machine learning and artificial intelligence are advancing fast.",
-    "Natural language processing techniques are widely adopted."
-]
-
-# Create Gradio interface
-iface = gr.Interface(
-    fn=text_to_speech,
-    inputs="text",
-    outputs="audio",
-    title="Fine-tuning TTS for English with a Focus on Technical Vocabulary Using SpeechT5",
-    description="Enter text with technical jargon for TTS conversion.",
-    examples=examples,
-    cache_examples=False  # Disable caching for now
-)
-
-# Launch interface
-iface.launch()
+
+    # Save the generated speech as a .wav file
+    output_file = "speech_output.wav"
+    sf.write(output_file, speech.numpy(), samplerate=16000)
+
+    return output_file
+
+# Step 5: Create Gradio interface
+iface = gr.Interface(fn=text_to_speech,
+                     inputs="text",
+                     outputs="audio",
+                     title="Text-to-Speech (TTS) Application",
+                     description="Enter text with technical jargon for TTS conversion.")
+
+# Step 6: Launch the app
+iface.launch(share=True)