app.py

import gradio as gr
import torch
import os
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
from datasets import load_dataset, Audio
import numpy as np
from speechbrain.inference import EncoderClassifier

# Load models and processor
processor = SpeechT5Processor.from_pretrained("microsoft/speecht5_tts")
model = SpeechT5ForTextToSpeech.from_pretrained("YOUR_FINE_TUNED_MODEL_PATH")
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")

# Load speaker encoder
device = "cuda" if torch.cuda.is_available() else "cpu"
speaker_model = EncoderClassifier.from_hparams(
    source="speechbrain/spkrec-xvect-voxceleb",
    run_opts={"device": device},
    savedir=os.path.join("/tmp", "speechbrain/spkrec-xvect-voxceleb")
)

# Load a sample from the dataset for speaker embedding
try:
    dataset = load_dataset("Yassmen/TTS_English_Technical_data", split="train")
    dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
    sample = dataset[0]
    speaker_embedding = create_speaker_embedding(sample['audio']['array'])
except Exception as e:
    print(f"Error loading dataset: {e}")
    # Use a random speaker embedding as fallback
    speaker_embedding = torch.randn(1, 512)

def create_speaker_embedding(waveform):
    with torch.no_grad():
        speaker_embeddings = speaker_model.encode_batch(torch.tensor(waveform))
        speaker_embeddings = torch.nn.functional.normalize(speaker_embeddings, dim=2)
        speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy()
    return speaker_embeddings

def text_to_speech(text):
    # Clean up text
    replacements = [
        ('$', 'dollar'), ('%', 'percent'), ('&', 'and'), ('*', 'asterick'),
        ('+', 'plus'), ('1', 'one'), ('2', 'two'), ('3', 'three'), ('4', 'four'),
        ('5', 'five'), ('6', 'six'), ('7', 'seven'), ('8', 'eight'), ('9', 'nine'),
        ('0', 'zero'), ('@', 'at'), ('\n', ' '), ('\xa0', ' '), (',', ' '),
        ('"', '"'), ('"', '"'),
    ]
    for src, dst in replacements:
        text = text.replace(src, dst)

    inputs = processor(text=text, return_tensors="pt")
    speech = model.generate_speech(inputs["input_ids"], speaker_embedding, vocoder=vocoder)
    return (16000, speech.numpy())

iface = gr.Interface(
    fn=text_to_speech,
    inputs="text",
    outputs="audio",
    title="Technical Text-to-Speech",
    description="Enter technical text to convert to speech. The model has been fine-tuned on technical data."
)

iface.launch()