import gradio as gr
import torch
from datasets import load_dataset
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
import soundfile as sf

# Load the fine-tuned model, processor, and vocoder
model_name = "microsoft/speecht5_tts"
processor = SpeechT5Processor.from_pretrained(model_name)
model = SpeechT5ForTextToSpeech.from_pretrained("emirhanbilgic/speecht5_finetuned_emirhan_tr")
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

# Load speaker embeddings (using the same as before, as it's not clear if a specific embedding is needed for the Turkish model)
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)


def text_to_speech(text):
    inputs = processor(text=text, return_tensors="pt")
    speech = model.generate_speech(inputs["input_ids"], speaker_embeddings, vocoder=vocoder)
    sf.write("output.wav", speech.numpy(), samplerate=16000)
    return "output.wav"

# Create Gradio interface
iface = gr.Interface(
    fn=text_to_speech,
    inputs=gr.Textbox(label="Enter Turkish text to convert to speech", value="Yapay zekayı seviyorum."),
    outputs=gr.Audio(label="Generated Speech"),
    title="Turkish SpeechT5 Text-to-Speech Demo",
    description="Enter Turkish text and listen to the generated speech using the fine-tuned SpeechT5 model."
)

# Launch the demo
iface.launch()