Update app.py

app.py

@@ -1,17 +1,39 @@
-from transformers import SpeechT5HifiGan
+import gradio as gr
+from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
+from datasets import load_dataset
 import torch
+import soundfile as sf
+import numpy as np
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
 
-# Load TTS model
+# Model
 model_name = "microsoft/speecht5_tts"
 processor = SpeechT5Processor.from_pretrained(model_name)
 model = SpeechT5ForTextToSpeech.from_pretrained(model_name).to(device)
-
-# Load default HiFi-GAN vocoder
 vocoder = SpeechT5HifiGan.from_pretrained(model_name)
 
-# Example speaker embedding
-# Hugging Face dataset example: cmu-arctic-xvectors
-# Here we can use 'matthijs/cmu-arctic-xvectors' speaker
-from datasets import load_dataset
+# Load a default speaker embedding
 dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="train[:1]")
 speaker_embedding = torch.tensor(dataset[0]["xvector"]).unsqueeze(0).to(device)
+
+# TTS function
+def text_to_speech(text):
+    inputs = processor(text=text, return_tensors="pt").to(device)
+    with torch.no_grad():
+        speech = model.generate_speech(inputs["input_ids"], speaker_embeddings=speaker_embedding)
+    
+    audio_np = speech.squeeze().cpu().numpy()
+    samplerate = processor.feature_extractor.sampling_rate
+    sf.write("output.wav", audio_np, samplerate)
+    return (audio_np, samplerate)
+
+iface = gr.Interface(
+    fn=text_to_speech,
+    inputs=gr.Textbox(lines=2, placeholder="Type your text here..."),
+    outputs=gr.Audio(type="numpy", autoplay=True),
+    title="SpeechT5 TTS Demo",
+    description="Type text and get speech output with default speaker voice."
+)
+
+iface.launch()