app.py

import gradio as gr
import tempfile
from TTS.api import TTS
from huggingface_hub import hf_hub_download
import torch

CUDA = torch.cuda.is_available()

REPO_ID = "ayymen/Coqui-TTS-Vits-shi"

my_title = "Tamazight Text-to-Speech"
my_description = "This model is based on [VITS](https://github.com/jaywalnut310/vits), thanks to 🐸 [Coqui.ai](https://coqui.ai/)." 

my_examples = [
  ["ⴰⵣⵓⵍ. ⵎⴰⵏⵣⴰⴽⵉⵏ?"],
  ["ⵡⴰ ⵜⴰⵎⵖⴰⵔⵜ ⵎⴰ ⴷ ⵓⴽⴰⵏ ⵜⵙⴽⵔⵜ?"],
  ["ⴳⵏ ⴰⴷ ⴰⴽ ⵉⵙⵙⴳⵏ ⵕⴱⴱⵉ ⵉⵜⵜⵓ ⴽ."],
  ["ⴰⵔⵔⴰⵡ ⵏ ⵍⵀⵎⵎ ⵢⵓⴽⵔ ⴰⵖ ⵉⵀⴷⵓⵎⵏ ⵏⵏⵖ!"]
]

my_inputs = [
  gr.Textbox(lines=5, label="Input Text"),
  gr.Audio(type="filepath", label="Speaker audio for voice cloning (optional)"),
  gr.Checkbox(label="Split Sentences (each sentence will be generated separately)", value=True)
]

my_outputs = gr.Audio(type="filepath", label="Output Audio", autoplay=True)

best_model_path = hf_hub_download(repo_id=REPO_ID, filename="best_model.pth") 
config_path = hf_hub_download(repo_id=REPO_ID, filename="config.json")

api = TTS(model_path=best_model_path, config_path=config_path).to("cuda" if CUDA else "cpu")

# load voice conversion model
api.load_vc_model_by_name("voice_conversion_models/multilingual/vctk/freevc24", gpu=CUDA)

def tts(text: str, speaker_wav: str = None, split_sentences: bool = True):
    # replace oov characters
    text = text.replace("\n", ". ")
    text = text.replace("(", ",")
    text = text.replace(")", ",")
    text = text.replace(";", ",")

    with tempfile.NamedTemporaryFile(suffix = ".wav", delete = False) as fp:
        if speaker_wav:
            api.tts_with_vc_to_file(text, speaker_wav=speaker_wav, file_path=fp.name, split_sentences=split_sentences)
        else:
            api.tts_to_file(text, file_path=fp.name, split_sentences=split_sentences)

    return fp.name 

iface = gr.Interface(
    fn=tts, 
    inputs=my_inputs, 
    outputs=my_outputs, 
    title=my_title, 
    description=my_description, 
    examples=my_examples,
    cache_examples=True
)
iface.launch()