app.py

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

model_id = "Sandiago21/speecht5_finetuned_google_fleurs_greek"  # update with your model id
# pipe = pipeline("automatic-speech-recognition", model=model_id)
model = SpeechT5ForTextToSpeech.from_pretrained(model_id)
vocoder = SpeechT5HifiGan.from_pretrained("microsoft/speecht5_hifigan")
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
speaker_embeddings = torch.tensor(embeddings_dataset[7440]["xvector"]).unsqueeze(0)

# checkpoint = "microsoft/speecht5_tts"
processor = SpeechT5Processor.from_pretrained(model_id)

replacements = [
    ("ου", "u"),
    ("αυ", "af"),
    ("ευ", "ef"),
    ("ει", "i"),
    ("οι", "i"),
    ("αι", "e"),
    ("ού", "u"),
    ("εί", "i"),
    ("οί", "i"),
    ("αί", "e"),
    ("Ά", "A"),
    ("Έ", "E"),
    ("Ή", "H"),
    ("Ί", "I"),
    ("Ό", "O"),
    ("Ύ", "Y"),
    ("Ώ", "O"),
    ("ΐ", "i"),
    ("Α", "A"),
    ("Β", "B"),
    ("Γ", "G"),
    ("Δ", "L"),
    ("Ε", "Ε"),
    ("Ζ", "Z"),
    ("Η", "I"),
    ("Θ", "Th"),
    ("Ι", "I"),
    ("Κ", "K"),
    ("Λ", "L"),
    ("Μ", "M"),
    ("Ν", "N"),
    ("Ξ", "Ks"),
    ("Ο", "O"),
    ("Π", "P"),
    ("Ρ", "R"),
    ("Σ", "S"),
    ("Τ", "T"),
    ("Υ", "Y"),
    ("Φ", "F"),
    ("Χ", "X"),
    ("Ω", "O"),
    ("ά", "a"),
    ("έ", "e"),
    ("ή", "i"),
    ("ί", "i"),
    ("α", "a"),
    ("β", "v"),
    ("γ", "g"),
    ("δ", "d"),
    ("ε", "e"),
    ("ζ", "z"),
    ("η", "i"),
    ("θ", "th"),
    ("ι", "i"),
    ("κ", "k"),
    ("λ", "l"),
    ("μ", "m"),
    ("ν", "n"),
    ("ξ", "ks"),
    ("ο", "o"),
    ("π", "p"),
    ("ρ", "r"),
    ("ς", "s"),
    ("σ", "s"),
    ("τ", "t"),
    ("υ", "i"),
    ("φ", "f"),
    ("χ", "h"),
    ("ψ", "ps"),
    ("ω", "o"),
    ("ϊ", "i"),
    ("ϋ", "i"),
    ("ό", "o"),
    ("ύ", "i"),
    ("ώ", "o"),
    ("í", "i"),
    ("õ", "o"),
    ("Ε", "E"),
    ("Ψ", "Ps"),
]


def cleanup_text(text):
    for src, dst in replacements:
        text = text.replace(src, dst)
    return text

def synthesize_speech(text):
    text = cleanup_text(text)
    inputs = processor(text=text, return_tensors="pt")

    speech = model.generate_speech(inputs["input_ids"], speaker_embeddings, vocoder=vocoder)

    return gr.Audio.update(value=(16000, speech.cpu().numpy()))

syntesize_speech_gradio = gr.Interface(
    synthesize_speech,
    inputs = gr.Textbox(label="Text", placeholder="Type something here..."),
    outputs=gr.Audio(),
    examples=["Έλαβαν χώρα μεγάλες διαδηλώσεις στην Πολωνία όταν εκείνη η χώρα υπέγραψε την acta που οδήγησε την κυβέρνηση της πολωνίας να αποφασίσει τη μη επικύρωση της συμφωνίας προς το παρόν"],
).launch()