app.py

import os
from huggingface_hub import snapshot_download
import gradio as gr
import spaces
from infer_rvc_python import BaseLoader
import random
import logging
import time
import soundfile as sf
from infer_rvc_python.main import download_manager, load_hu_bert, Config
import zipfile
import edge_tts
import asyncio
import librosa
import traceback
import soundfile as sf
from pedalboard import Pedalboard, Reverb, Compressor, HighpassFilter
from pedalboard.io import AudioFile
from pydub import AudioSegment
import noisereduce as nr
import numpy as np
import urllib.request
import shutil
import threading
import argparse
import sys
import torch
import fairseq

# =====================================================================
# កូដទាញយកម៉ូដែលសំឡេងពី Repository
# =====================================================================
print("📥 កំពុងទាញយកម៉ូដែលសំឡេងទាំងអស់ពី Repository... សូមរង់ចាំ!")
REPO_ID = "Tha456/Khmer-Voice-Models"
LOCAL_MODELS_DIR = snapshot_download(repo_id=REPO_ID, local_dir="my_models")
print(f"✅ ទាញយកជោគជ័យ! ឯកសារទាំងអស់ស្ថិតនៅក្នុង Folder: {LOCAL_MODELS_DIR}")

# បង្កើតអថេរផ្លូវ (Path) សម្រាប់កូដខាងក្រោមយកទៅប្រើ
nita_pth = os.path.join(LOCAL_MODELS_DIR, "nita_female.pth")
nita_index = os.path.join(LOCAL_MODELS_DIR, "nita_female.index")

nimol_pth = os.path.join(LOCAL_MODELS_DIR, "nimol_famale.pth")
nimol_index = os.path.join(LOCAL_MODELS_DIR, "nimol_famale.index")

saman_pth = os.path.join(LOCAL_MODELS_DIR, "saman_male.pth")
saman_index = os.path.join(LOCAL_MODELS_DIR, "saman_male.index")

sana_pth = os.path.join(LOCAL_MODELS_DIR, "sana_femal.pth")
sana_index = os.path.join(LOCAL_MODELS_DIR, "sana_femal.index")

sovanna_pth = os.path.join(LOCAL_MODELS_DIR, "sovanna_male.pth")
sovanna_index = os.path.join(LOCAL_MODELS_DIR, "sovanna_male.index")

# បង្កើត Dictionary សម្រាប់គ្រប់គ្រងម៉ូដែលសំឡេងខ្មែរ
KHMER_VOICES = {
    "នីតា (Nita - Female)": {"pth": nita_pth, "index": nita_index},
    "និមល (Nimol - Female)": {"pth": nimol_pth, "index": nimol_index},
    "សាម៉ន (Saman - Male)": {"pth": saman_pth, "index": saman_index},
    "សាណា (Sana - Female)": {"pth": sana_pth, "index": sana_index},
    "សុវណ្ណា (Sovanna - Male)": {"pth": sovanna_pth, "index": sovanna_index},
}
# =====================================================================

parser = argparse.ArgumentParser(description="Run the app with optional sharing")
parser.add_argument(
    '--share',
    action='store_true',
    help='Enable sharing mode'
)
parser.add_argument(
    '--theme',
    type=str,
    default="aliabid94/new-theme",
    help='Set the theme (default: aliabid94/new-theme)'
)
args = parser.parse_args()

IS_COLAB = True if ('google.colab' in sys.modules or args.share) else False
IS_ZERO_GPU = os.getenv("SPACES_ZERO_GPU")

logging.getLogger("infer_rvc_python").setLevel(logging.ERROR)

torch.serialization.add_safe_globals([fairseq.data.dictionary.Dictionary])
converter = BaseLoader(only_cpu=False, hubert_path=None, rmvpe_path=None)
converter.hu_bert_model = load_hu_bert(Config(only_cpu=False), converter.hubert_path)

test_model = "https://huggingface.co/sail-rvc/Aldeano_Minecraft__RVC_V2_-_500_Epochs_/resolve/main/model.pth?download=true, https://huggingface.co/sail-rvc/Aldeano_Minecraft__RVC_V2_-_500_Epochs_/resolve/main/model.index?download=true"
test_names = ["model.pth", "model.index"]

for url, filename in zip(test_model.split(", "), test_names):
    try:
        download_manager(
            url=url,
            path=".",
            extension="",
            overwrite=False,
            progress=True,
        )
        if not os.path.isfile(filename):
            raise FileNotFoundError
    except Exception:
        with open(filename, "wb") as f:
            pass

title = "<center><strong><font size='7'>RVC⚡ZERO Khmer Edition</font></strong></center>"
description = "This demo is provided for educational and research purposes only." if IS_ZERO_GPU else ""
RESOURCES = "- You can also try `RVC⚡ZERO` in Colab’s free tier."
theme = args.theme
delete_cache_time = (3200, 3200) if IS_ZERO_GPU else (86400, 86400)

PITCH_ALGO_OPT = ["pm", "harvest", "crepe", "rmvpe", "rmvpe+"]

async def get_voices_list(proxy=None):
    from edge_tts import list_voices
    voices = await list_voices(proxy=proxy)
    voices = sorted(voices, key=lambda voice: voice["ShortName"])
    table = [
        {
            "ShortName": voice["ShortName"],
            "Gender": voice["Gender"],
            "ContentCategories": ", ".join(voice["VoiceTag"]["ContentCategories"]),
            "VoicePersonalities": ", ".join(voice["VoiceTag"]["VoicePersonalities"]),
            "FriendlyName": voice["FriendlyName"],
        }
        for voice in voices
    ]
    return table

def find_files(directory):
    file_paths = []
    for filename in os.listdir(directory):
        if filename.endswith('.pth') or filename.endswith('.zip') or filename.endswith('.index'):
            file_paths.append(os.path.join(directory, filename))
    return file_paths

def unzip_in_folder(my_zip, my_dir):
    with zipfile.ZipFile(my_zip) as zip:
        for zip_info in zip.infolist():
            if zip_info.is_dir():
                continue
            zip_info.filename = os.path.basename(zip_info.filename)
            zip.extract(zip_info, my_dir)

def find_my_model(a_, b_):
    if a_ is None or a_.endswith(".pth"):
        return a_, b_
    txt_files = []
    for base_file in [a_, b_]:
        if base_file is not None and base_file.endswith(".txt"):
            txt_files.append(base_file)
    directory = os.path.dirname(a_)
    for txt in txt_files:
        with open(txt, 'r') as file:
            first_line = file.readline()
        download_manager(url=first_line.strip(), path=directory, extension="")
    for f in find_files(directory):
        if f.endswith(".zip"):
            unzip_in_folder(f, directory)
    model = None
    index = None
    end_files = find_files(directory)
    for ff in end_files:
        if ff.endswith(".pth"):
            model = os.path.join(directory, ff)
            gr.Info(f"Model found: {ff}")
        if ff.endswith(".index"):
            index = os.path.join(directory, ff)
            gr.Info(f"Index found: {ff}")
    if not model:
        gr.Error(f"Model not found in: {end_files}")
    if not index:
        gr.Warning("Index not found")
    return model, index

def ensure_valid_file(url):
    if "huggingface" not in url:
        raise ValueError("Only downloads from Hugging Face are allowed")
    try:
        request = urllib.request.Request(url, method="HEAD")
        with urllib.request.urlopen(request) as response:
            content_length = response.headers.get("Content-Length")
        if content_length is None:
            raise ValueError("No Content-Length header found")
        file_size = int(content_length)
        if file_size > 900000000 and IS_ZERO_GPU:
            raise ValueError("The file is too large. Max allowed is 900 MB.")
        return file_size
    except Exception as e:
        raise e

def clear_files(directory):
    time.sleep(15)
    print(f"Clearing files: {directory}.")
    shutil.rmtree(directory)

def get_my_model(url_data, progress=gr.Progress(track_tqdm=True)):
    if not url_data:
        return None, None
    if "," in url_data:
        a_, b_ = url_data.split(",")
        a_, b_ = a_.strip().replace("/blob/", "/resolve/"), b_.strip().replace("/blob/", "/resolve/")
    else:
        a_, b_ = url_data.strip().replace("/blob/", "/resolve/"), None
    out_dir = "downloads"
    folder_download = str(random.randint(1000, 9999))
    directory = os.path.join(out_dir, folder_download)
    os.makedirs(directory, exist_ok=True)
    try:
        valid_url = [a_] if not b_ else [a_, b_]
        for link in valid_url:
            ensure_valid_file(link)
            download_manager(url=link, path=directory, extension="")
        for f in find_files(directory):
            if f.endswith(".zip"):
                unzip_in_folder(f, directory)
        model = None
        index = None
        end_files = find_files(directory)
        for ff in end_files:
            if ff.endswith(".pth"):
                model = ff
                gr.Info(f"Model found: {ff}")
            if ff.endswith(".index"):
                index = ff
                gr.Info(f"Index found: {ff}")
        if not model:
            raise ValueError(f"Model not found in: {end_files}")
        if not index:
            gr.Warning("Index not found")
        else:
            index = os.path.abspath(index)
        return os.path.abspath(model), index
    except Exception as e:
        raise e
    finally:
        t = threading.Thread(target=clear_files, args=(directory,))
        t.start()

def add_audio_effects(audio_list, type_output):
    result = []
    for audio_path in audio_list:
        try:
            output_path = f'{os.path.splitext(audio_path)[0]}_effects.{type_output}'
            board = Pedalboard(
                [
                    HighpassFilter(),
                    Compressor(ratio=4, threshold_db=-15),
                    Reverb(room_size=0.10, dry_level=0.8, wet_level=0.2, damping=0.7)
                ]
            )
            temp_wav = f'{os.path.splitext(audio_path)[0]}_temp.wav'
            with AudioFile(audio_path) as f:
                with AudioFile(temp_wav, 'w', f.samplerate, f.num_channels) as o:
                    while f.tell() < f.frames:
                        chunk = f.read(int(f.samplerate))
                        effected = board(chunk, f.samplerate, reset=False)
                        o.write(effected)
            audio_seg = AudioSegment.from_file(temp_wav, format=type_output)
            audio_seg.export(output_path, format=type_output, bitrate=("320k" if type_output == "mp3" else None))
            os.remove(temp_wav)
            result.append(output_path)
        except Exception as e:
            traceback.print_exc()
            result.append(audio_path)
    return result

def apply_noisereduce(audio_list, type_output):
    result = []
    for audio_path in audio_list:
        out_path = f"{os.path.splitext(audio_path)[0]}_noisereduce.{type_output}"
        try:
            audio = AudioSegment.from_file(audio_path)
            samples = np.array(audio.get_array_of_samples())
            reduced_noise = nr.reduce_noise(samples, sr=audio.frame_rate, prop_decrease=0.6)
            reduced_audio = AudioSegment(
                reduced_noise.tobytes(), 
                frame_rate=audio.frame_rate, 
                sample_width=audio.sample_width,
                channels=audio.channels
            )
            reduced_audio.export(out_path, format=type_output, bitrate=("320k" if type_output == "mp3" else None))
            result.append(out_path)
        except Exception as e:
            traceback.print_exc()
            result.append(audio_path)
    return result

@spaces.GPU()
def convert_now(audio_files, random_tag, converter, type_output, steps):
    for step in range(steps):
        audio_files = converter(
            audio_files,
            random_tag,
            overwrite=False,
            parallel_workers=(2 if IS_COLAB else 8),
            type_output=type_output,
        )
    return audio_files

# --- កែប្រែត្រង់ចំណុចនេះ៖ បន្ថែមអថេរ voice_select ដើម្បីចាប់យកម៉ូដែលស្វ័យប្រវត្តិ ---
def run(
    audio_files,
    file_m,
    pitch_alg,
    pitch_lvl,
    file_index,
    index_inf,
    r_m_f,
    e_r,
    c_b_p,
    active_noise_reduce,
    audio_effects,
    type_output,
    steps,
    voice_select="Upload Custom Model", 
):
    # ប្រសិនបើអ្នកប្រើរើសសំឡេងខ្មែរដែលមានស្រាប់ វានឹងកំណត់យកហ្វាយពី Repository ភ្លាម
    if voice_select in KHMER_VOICES:
        file_m = KHMER_VOICES[voice_select]["pth"]
        file_index = KHMER_VOICES[voice_select]["index"]

    if not audio_files:
        raise ValueError("The audio pls")

    if isinstance(audio_files, str):
        audio_files = [audio_files]

    try:
        duration_base = librosa.get_duration(filename=audio_files[0])
        print("Duration:", duration_base)
    except Exception as e:
        print(e)

    if file_m is not None and file_m.endswith(".txt"):
        file_m, file_index = find_my_model(file_m, file_index)
        print(file_m, file_index)

    random_tag = "USER_"+str(random.randint(10000000, 99999999))

    converter.apply_conf(
        tag=random_tag,
        file_model=file_m,
        pitch_algo=pitch_alg,
        pitch_lvl=pitch_lvl,
        file_index=file_index,
        index_influence=index_inf,
        respiration_median_filtering=r_m_f,
        envelope_ratio=e_r,
        consonant_breath_protection=c_b_p,
        resample_sr=0,
    )
    time.sleep(0.1)

    result = convert_now(audio_files, random_tag, converter, type_output, steps)

    if active_noise_reduce:
        result = apply_noisereduce(result, type_output)

    if audio_effects:
        result = add_audio_effects(result, type_output)

    return result

def audio_conf():
    return gr.File(label="Audio files", file_count="multiple", type="filepath", container=True)

def model_conf():
    return gr.File(label="Model file (.pth)", type="filepath", height=130, visible=False) # លាក់លំនាំដើម

def pitch_algo_conf():
    return gr.Dropdown(PITCH_ALGO_OPT, value=PITCH_ALGO_OPT[4], label="Pitch algorithm", visible=True, interactive=True)

def pitch_lvl_conf():
    return gr.Slider(label="Pitch level", minimum=-24, maximum=24, step=1, value=0, visible=True, interactive=True)

def index_conf():
    return gr.File(label="Index file (.index)", type="filepath", height=130, visible=False) # លាក់លំនាំដើម

def index_inf_conf():
    return gr.Slider(minimum=0, maximum=1, label="Index influence", value=0.75)

def respiration_filter_conf():
    return gr.Slider(minimum=0, maximum=7, label="Respiration median filtering", value=3, step=1, interactive=True)

def envelope_ratio_conf():
    return gr.Slider(minimum=0, maximum=1, label="Envelope ratio", value=0.25, interactive=True)

def consonant_protec_conf():
    return gr.Slider(minimum=0, maximum=0.5, label="Consonant breath protection", value=0.5, interactive=True)

def button_conf():
    return gr.Button("Inference", variant="primary")

def output_conf():
    return gr.File(label="Result", file_count="multiple", interactive=False)

def active_tts_conf():
    return gr.Checkbox(False, label="TTS", container=False)

def tts_voice_conf():
    return gr.Dropdown(label="tts voice", choices=voices, visible=False, value="en-US-EmmaMultilingualNeural-Female")

def tts_text_conf():
    return gr.Textbox(value="", placeholder="Write the text here...", label="Text", visible=False, lines=3)

def tts_button_conf():
    return gr.Button("Process TTS", variant="secondary", visible=False)

def tts_play_conf():
    return gr.Checkbox(False, label="Play", container=False, visible=False)

def sound_gui():
    return gr.Audio(value=None, type="filepath", autoplay=True, visible=True, interactive=False, elem_id="audio_tts")

def steps_conf():
    return gr.Slider(minimum=1, maximum=3, label="Steps", value=1, step=1, interactive=True)

def format_output_gui():
    return gr.Dropdown(label="Format output:", choices=["wav", "mp3", "flac"], value="wav")

def denoise_conf():
    return gr.Checkbox(False, label="Denoise", container=False, visible=True)

def effects_conf():
    return gr.Checkbox(False, label="Reverb", container=False, visible=True)

def infer_tts_audio(tts_voice, tts_text, play_tts):
    out_dir = "output"
    folder_tts = "USER_"+str(random.randint(10000, 99999))
    os.makedirs(out_dir, exist_ok=True)
    os.makedirs(os.path.join(out_dir, folder_tts), exist_ok=True)
    out_path = os.path.join(out_dir, folder_tts, "tts.mp3")
    asyncio.run(edge_tts.Communicate(tts_text, "-".join(tts_voice.split('-')[:-1])).save(out_path))
    if play_tts:
        return [out_path], out_path
    return [out_path], None

def show_components_tts(value_active):
    return gr.update(visible=value_active), gr.update(visible=value_active), gr.update(visible=value_active), gr.update(visible=value_active)

def down_active_conf():
    return gr.Checkbox(False, label="URL-to-Model", container=False)

def down_url_conf():
    return gr.Textbox(value="", placeholder="Write the url here...", label="Enter URL", visible=False, lines=1)

def down_button_conf():
    return gr.Button("Process", variant="secondary", visible=False)

def show_components_down(value_active):
    return gr.update(visible=value_active), gr.update(visible=value_active), gr.update(visible=value_active)

# បង្កើតមុខងារកែប្រែភាពមើលឃើញរបស់ File Upload
def update_model_visibility(voice_choice):
    if voice_choice == "Upload Custom Model":
        return gr.update(visible=True), gr.update(visible=True)
    return gr.update(visible=False), gr.update(visible=False)

CSS = """
#audio_tts { visibility: hidden; height: 0px; width: 0px; max-width: 0px; max-height: 0px; }
"""

def get_gui(theme):
    with gr.Blocks(theme=theme, css=CSS, fill_width=True, fill_height=False, delete_cache=delete_cache_time) as app:
        gr.Markdown(title)
        gr.Markdown(description)

        active_tts = active_tts_conf()
        with gr.Row():
            with gr.Column(scale=1):
                tts_text = tts_text_conf()
            with gr.Column(scale=2):
                with gr.Row():
                    with gr.Column():
                        with gr.Row():
                            tts_voice = tts_voice_conf()
                            tts_active_play = tts_play_conf()
                tts_button = tts_button_conf()
                tts_play = sound_gui()

        active_tts.change(
            fn=show_components_tts,
            inputs=[active_tts],
            outputs=[tts_voice, tts_text, tts_button, tts_active_play],
        )

        # --- កែប្រែត្រង់ចំណុចនេះ៖ បន្ថែម Dropdown ជ្រើសរើសសំឡេងខ្មែរ ---
        voice_select = gr.Dropdown(
            choices=list(KHMER_VOICES.keys()) + ["Upload Custom Model"],
            value="នីតា (Nita - Female)",
            label="🎙️ ជ្រើសរើសម៉ូដែលសំឡេងខ្មែរ (Select Khmer Voice Model)",
        )

        aud = audio_conf()

        tts_button.click(
            fn=infer_tts_audio,
            inputs=[tts_voice, tts_text, tts_active_play],
            outputs=[aud, tts_play],
        )

        down_active_gui = down_active_conf()
        down_info = gr.Markdown(
            f"Provide a link to a zip file...",
            visible=False
        )
        with gr.Row():
            with gr.Column(scale=3):
                down_url_gui = down_url_conf()
            with gr.Column(scale=1):
                down_button_gui = down_button_conf()

        with gr.Column():
            with gr.Row():
                model = model_conf()
                indx = index_conf()

        # កំណត់ឱ្យលាក់ ឬបង្ហាញផ្ទាំង Upload ទៅតាមការរើសសំឡេង
        voice_select.change(
            fn=update_model_visibility,
            inputs=[voice_select],
            outputs=[model, indx]
        )

        down_active_gui.change(
            show_components_down,
            [down_active_gui],
            [down_info, down_url_gui, down_button_gui]
        )

        down_button_gui.click(
            get_my_model,
            [down_url_gui],
            [model, indx]
        )

        with gr.Accordion(label="Advanced settings", open=False):
            algo = pitch_algo_conf()
            algo_lvl = pitch_lvl_conf()
            indx_inf = index_inf_conf()
            res_fc = respiration_filter_conf()
            envel_r = envelope_ratio_conf()
            const = consonant_protec_conf()
            steps_gui = steps_conf()
            format_out = format_output_gui()
            with gr.Row():
                with gr.Column():
                    with gr.Row():
                        denoise_gui = denoise_conf()
                        effects_gui = effects_conf()
        
        button_base = button_conf()
        output_base = output_conf()

        # បញ្ចូល voice_select ទៅក្នុងប្រព័ន្ធរត់កូដ
        button_base.click(
            run,
            inputs=[
                aud,
                model,
                algo,
                algo_lvl,
                indx,
                indx_inf,
                res_fc,
                envel_r,
                const,
                denoise_gui,
                effects_gui,
                format_out,
                steps_gui,
                voice_select, # ថែមត្រង់នេះ
            ],
            outputs=[output_base],
        )

        gr.Examples(
            examples=[
                [["./test.ogg"], "./model.pth", "rmvpe+", 0, "./model.index", 0.75, 3, 0.25, 0.50],
            ],
            fn=run,
            inputs=[aud, model, algo, algo_lvl, indx, indx_inf, res_fc, envel_r, const],
            outputs=[output_base],
            cache_examples=False,
        )
        gr.Markdown(RESOURCES)

    return app

if __name__ == "__main__":
    tts_voice_list = asyncio.new_event_loop().run_until_complete(get_voices_list(proxy=None))
    voices = sorted([
        (" - ".join(reversed(v["FriendlyName"].split("-"))).replace("Microsoft ", "").replace("Online (Natural)", f"({v['Gender']})").strip(), f"{v['ShortName']}-{v['Gender']}")
        for v in tts_voice_list
    ])

    app = get_gui(theme)
    app.queue(default_concurrency_limit=40)
    app.launch(
        max_threads=40,
        share=IS_COLAB,
        show_error=True,
        quiet=False,
        debug=IS_COLAB,
        ssr_mode=False,
    )