app_gradio.py

import gradio as gr
import os
import json
import math
import torch
from torch import nn
from torch.nn import functional as F
import librosa
import argparse
import librosa.display
import matplotlib.pyplot as plt


import commons
import utils
from data_utils import TextAudioLoader, TextAudioCollate, TextAudioSpeakerLoader, TextAudioSpeakerCollate
#from model_old_mel_style import SynthesizerTrn
from models_mel_style import SynthesizerTrn
from text.symbols import symbols
from text import text_to_sequence
from mel_processing import spectrogram_torch, spec_to_mel_torch
from scipy.io.wavfile import write

# Thư mục chứa các file wav
AUDIO_DIR = "wav/wav_1"

def list_wav_files():
    return [f for f in os.listdir(AUDIO_DIR) if f.endswith(".wav")]

# Trả về đường dẫn file wav được chọn
def get_audio_file(file_name):
    file_path = os.path.join(AUDIO_DIR, file_name)
    return file_path

def get_text(text, hps):
    text_norm = text_to_sequence(text, hps.data.text_cleaners)
    if hps.data.add_blank:
        text_norm = commons.intersperse(text_norm, 0)
    text_norm = torch.LongTensor(text_norm)
    return text_norm

# Tạo giọng nói bằng mô hình
def generate_voice(prompt_text, ref_audio_filename):
    import argparse
    class Args:
        checkpoint_path = "logs/large_audio/G_504000.pth"
        config = "configs/vn_base.json"
        save_path = "infer_result/"
        ref_audio = os.path.join("wav/wav_1", ref_audio_filename)
        text = prompt_text
    args = Args()

    hps = utils.get_hparams_from_file(args.config)
    net_g = SynthesizerTrn(
        len(symbols),
        hps.data.filter_length // 2 + 1,
        hps.train.segment_size // hps.data.hop_length,
        n_speakers=0,
        **hps.model
    )
    _ = net_g.eval()
    _ = utils.load_checkpoint(args.checkpoint_path, net_g, None)

    audio, _ = librosa.load(args.ref_audio, sr=hps.data.sampling_rate)
    audio = torch.from_numpy(audio).unsqueeze(0)
    spec = spectrogram_torch(audio, hps.data.filter_length, hps.data.sampling_rate,
                             hps.data.hop_length, hps.data.win_length, center=False)
    spec = torch.squeeze(spec, 0)
    mel = spec_to_mel_torch(spec, hps.data.filter_length, hps.data.n_mel_channels,
                            hps.data.sampling_rate, hps.data.mel_fmin, hps.data.mel_fmax)

    stn_tst = get_text(args.text, hps)
    x_tst = stn_tst.unsqueeze(0)
    x_tst_lengths = torch.LongTensor([stn_tst.size(0)])
    sid = torch.LongTensor([4])

    with torch.no_grad():
        audio_gen = net_g.infer(x_tst, x_tst_lengths, mel.unsqueeze(0),
                                sid=None, noise_scale=0.1,
                                noise_scale_w=0.1, length_scale=1.1)[0][0, 0].data.cpu().float().numpy()

        os.makedirs(args.save_path, exist_ok=True)
        output_file = os.path.join(args.save_path, f'test_{str(len(os.listdir(args.save_path)))}.wav')
        write(output_file, hps.data.sampling_rate, audio_gen)

    return output_file

# with gr.Blocks() as demo:
#     gr.Markdown("<center># <h1>Demo Model Text to Speech</h1></center>")

#     prompt = gr.Textbox(label="Prompt", placeholder="Type somethrgs.ing here...")

#     wav_files = sorted(list_wav_files())
#     if not wav_files:
#         gr.Markdown("⚠️ Không tìm thấy file .wav nào trong thư mục!")

#     gr.Markdown("## 🎧 Chọn và nghe file âm thanh gốc")

#     with gr.Row():
#         file_dropdown = gr.Dropdown(choices=wav_files, label="Chọn file WAV")
#         audio_output = gr.Audio(type="filepath", label="Nghe tại đây")

#     file_dropdown.change(fn=get_audio_file, inputs=file_dropdown, outputs=audio_output)

#     generate_button = gr.Button("Generate Voice")

#     generated_audio_output = gr.Audio(type="filepath", label="🔊 Kết quả sinh giọng nói")
#     generate_button.click(fn=generate_voice, inputs=[prompt, file_dropdown], outputs=generated_audio_output)

with gr.Blocks() as demo:
    gr.Markdown("<center># <h1>Demo Model Text to Speech</h1></center>")

    prompt = gr.Textbox(label="Prompt", placeholder="Type something here...")

    gr.Markdown("## 🎧 Chọn hoặc ghi âm giọng nói tham chiếu")

    with gr.Tab("📁 Chọn từ file"):
        wav_files = sorted(list_wav_files())
        file_dropdown = gr.Dropdown(choices=wav_files, label="Chọn file WAV có sẵn")
        audio_output = gr.Audio(type="filepath", label="Nghe tại đây")
        file_dropdown.change(fn=get_audio_file, inputs=file_dropdown, outputs=audio_output)

    with gr.Tab("🎙️ Ghi âm mới"):
        recorded_audio = gr.Audio(label="Ghi âm hoặc chọn file", type="filepath")

    # Nút sinh giọng nói
    generate_button = gr.Button("Generate Voice")
    generated_audio_output = gr.Audio(type="filepath", label="🔊 Kết quả sinh giọng nói")

    def process_inputs(prompt_text, file_choice, recorded_path):
        # Nếu người dùng có file ghi âm -> lưu tạm và dùng
        if recorded_path is not None:
            filename = f"user_recording_{len(os.listdir(AUDIO_DIR))}.wav"
            saved_path = os.path.join(AUDIO_DIR, filename)
            os.rename(recorded_path, saved_path)
            ref_file = filename
        elif file_choice:
            ref_file = file_choice
        else:
            raise gr.Error("Bạn cần chọn hoặc ghi âm một file giọng nói.")

        return generate_voice(prompt_text, ref_file)

    generate_button.click(
        fn=process_inputs,
        inputs=[prompt, file_dropdown, recorded_audio],
        outputs=generated_audio_output
    )

demo.launch()