app.py

import csv
import datetime
import os
import re
import time
import uuid
from io import StringIO

import gradio as gr
import torch
import torchaudio
from huggingface_hub import HfApi, hf_hub_download, snapshot_download
from TTS.tts.configs.xtts_config import XttsConfig
from TTS.tts.models.xtts import Xtts
from vinorm import TTSnorm

# Initialize Hugging Face API
HF_TOKEN = os.environ.get("HF_TOKEN")
api = HfApi(token=HF_TOKEN)

# Download model files if not already downloaded
print("Downloading viXTTS model files if not already present...")
checkpoint_dir = "model/"
repo_id = "capleaf/viXTTS"
use_deepspeed = False

os.makedirs(checkpoint_dir, exist_ok=True)

required_files = ["model.pth", "config.json", "vocab.json", "speakers_xtts.pth"]
files_in_dir = os.listdir(checkpoint_dir)
if not all(file in files_in_dir for file in required_files):
    snapshot_download(
        repo_id=repo_id,
        repo_type="model",
        local_dir=checkpoint_dir,
    )
    hf_hub_download(
        repo_id="coqui/XTTS-v2",
        filename="speakers_xtts.pth",
        local_dir=checkpoint_dir,
    )

# Load model configuration and initialize model
xtts_config = os.path.join(checkpoint_dir, "config.json")
config = XttsConfig()
config.load_json(xtts_config)
MODEL = Xtts.init_from_config(config)
MODEL.load_checkpoint(
    config, checkpoint_dir=checkpoint_dir, use_deepspeed=use_deepspeed
)
if torch.cuda.is_available():
    MODEL.cuda()

# Supported languages
supported_languages = config.languages
if "vi" not in supported_languages:
    supported_languages.append("vi")


def normalize_vietnamese_text(text):
    text = (
        TTSnorm(text, unknown=False, lower=False, rule=True)
        .replace("..", ".")
        .replace("!.", "!")
        .replace("?.", "?")
        .replace(" .", ".")
        .replace(" ,", ",")
        .replace('"', "")
        .replace("'", "")
        .replace("AI", "Ây Ai")
        .replace("A.I", "Ây Ai")
    )
    return text


def calculate_keep_len(text, lang):
    if lang in ["ja", "zh-cn"]:
        return -1

    word_count = len(text.split())
    num_punct = text.count(".") + text.count("!") + text.count("?") + text.count(",")

    if word_count < 5:
        return 15000 * word_count + 2000 * num_punct
    elif word_count < 10:
        return 13000 * word_count + 2000 * num_punct
    return -1


def predict(prompt, language, audio_file_pth, normalize_text=True):
    if language not in supported_languages:
        metrics_text = gr.Warning(
            f"Language {language} is not supported. Please choose from the dropdown."
        )
        return None, metrics_text

    if len(prompt) < 2:
        metrics_text = gr.Warning("Please provide a longer prompt text.")
        return None, metrics_text

    try:
        metrics_text = ""
        t_latent = time.time()

        try:
            gpt_cond_latent, speaker_embedding = MODEL.get_conditioning_latents(
                audio_path=audio_file_pth,
                gpt_cond_len=30,
                gpt_cond_chunk_len=4,
                max_ref_length=60,
            )
        except Exception as e:
            print("Speaker encoding error:", str(e))
            metrics_text = gr.Warning("Error with reference audio.")
            return None, metrics_text

        prompt = re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)", r"\1 \2", prompt)

        if normalize_text and language == "vi":
            prompt = normalize_vietnamese_text(prompt)

        print("Generating new audio...")
        t0 = time.time()
        out = MODEL.inference(
            prompt,
            language,
            gpt_cond_latent,
            speaker_embedding,
            repetition_penalty=5.0,
            temperature=0.75,
            enable_text_splitting=True,
        )
        inference_time = time.time() - t0
        metrics_text += f"Time to generate audio: {round(inference_time * 1000)} ms\n"
        real_time_factor = (time.time() - t0) / out["wav"].shape[-1] * 24000
        metrics_text += f"Real-time factor (RTF): {real_time_factor:.2f}\n"

        keep_len = calculate_keep_len(prompt, language)
        out["wav"] = out["wav"][:keep_len]

        torchaudio.save("output.wav", torch.tensor(out["wav"]).unsqueeze(0), 24000)

    except RuntimeError as e:
        print("RuntimeError:", str(e))
        metrics_text = gr.Warning("An error occurred during processing.")
        return None, metrics_text

    return "output.wav", metrics_text


title = "viXTTS Demo"

with gr.Blocks(analytics_enabled=False) as demo:
    with gr.Row():
        with gr.Column():
            gr.Markdown("## viXTTS Demo")
        with gr.Column():
            pass

    with gr.Row():
        with gr.Column():
            input_text_gr = gr.Textbox(
                label="Text Prompt",
                info="One or two sentences at a time is better. Up to 200 text characters.",
                value="Xin chào, tôi là một mô hình chuyển đổi văn bản thành giọng nói tiếng Việt",
            )
            language_gr = gr.Dropdown(
                label="Language",
                info="Select an output language for the synthesised speech",
                choices=supported_languages,
                value="vi",
            )
            normalize_text = gr.Checkbox(
                label="Normalize Vietnamese Text",
                info="Normalize Vietnamese Text",
                value=True,
            )
            ref_gr = gr.Audio(
                label="Reference Audio",
                info="Click on the ✎ button to upload your own target speaker audio",
                type="filepath",
                value="model/samples/nu-luu-loat.wav",
            )
            tts_button = gr.Button("Send", elem_id="send-btn", visible=True)

        with gr.Column():
            audio_gr = gr.Audio(label="Synthesised Audio", autoplay=True)
            out_text_gr = gr.Textbox(label="Metrics")

    tts_button.click(
        predict,
        [input_text_gr, language_gr, ref_gr, normalize_text],
        outputs=[audio_gr, out_text_gr],
        api_name="predict",
    )

demo.queue()
demo.launch(debug=True, show_api=True)