app.py

import gradio as gr
import requests
import random
import os
import zipfile 
import librosa
import time
from infer_rvc_python import BaseLoader
from pydub import AudioSegment
from tts_voice import tts_order_voice
import edge_tts
import tempfile
from audio_separator.separator import Separator
import model_handler
import psutil
import cpuinfo

language_dict = tts_order_voice

async def text_to_speech_edge(text, language_code):
    voice = language_dict[language_code]
    communicate = edge_tts.Communicate(text, voice)
    with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
        tmp_path = tmp_file.name

    await communicate.save(tmp_path)

    return tmp_path

try:
    import spaces
    spaces_status = True
except ImportError:
    spaces_status = False

separator = Separator()
converter = BaseLoader(only_cpu=False, hubert_path=None, rmvpe_path=None)

global pth_file
global index_file

pth_file = "model.pth"
index_file = "model.index"

#CONFIGS
TEMP_DIR = "temp"
MODEL_PREFIX = "model"
PITCH_ALGO_OPT = [
    "pm",
    "harvest",
    "crepe",
    "rmvpe",
    "rmvpe+",
]
UVR_5_MODELS = [
    {"model_name": "BS-Roformer-Viperx-1297", "checkpoint": "model_bs_roformer_ep_317_sdr_12.9755.ckpt"},
    {"model_name": "MDX23C-InstVoc HQ 2", "checkpoint": "MDX23C-8KFFT-InstVoc_HQ_2.ckpt"},
    {"model_name": "Kim Vocal 2", "checkpoint": "Kim_Vocal_2.onnx"},
    {"model_name": "5_HP-Karaoke", "checkpoint": "5_HP-Karaoke-UVR.pth"},
    {"model_name": "UVR-DeNoise by FoxJoy", "checkpoint": "UVR-DeNoise.pth"},
    {"model_name": "UVR-DeEcho-DeReverb by FoxJoy", "checkpoint": "UVR-DeEcho-DeReverb.pth"},
]
MODELS = [
    {"model": "model.pth", "index": "model.index", "model_name": "Test Model"},
]

os.makedirs(TEMP_DIR, exist_ok=True)

def unzip_file(file):
    filename = os.path.basename(file).split(".")[0] 
    with zipfile.ZipFile(file, 'r') as zip_ref:
        zip_ref.extractall(os.path.join(TEMP_DIR, filename)) 
    return True
    

def progress_bar(total, current):
    return "[" + "=" * int(current / total * 20) + ">" + " " * (20 - int(current / total * 20)) + "] " + str(int(current / total * 100)) + "%"

def contains_bad_word(text, bad_words):
    text_lower = text.lower()
    for word in bad_words:
        if word.lower() in text_lower:
            return True
    return False

bad_words = ['puttana', 'whore', 'badword3', 'badword4']

class BadWordError(Exception):
    def __init__(self, msg):
        super().__init__(msg)
        self.word = word

def download_from_url(url, name=None):
    if name is None:
        raise ValueError("The model name must be provided")
    if "/blob/" in url:
        url = url.replace("/blob/", "/resolve/") 
    if "huggingface" not in url:
        return ["The URL must be from huggingface", "Failed", "Failed"]
    if contains_bad_word(url, bad_words):
        return BadWordError("The file url has a bad word.")
    if contains_bad_word(name, bad_words):
        return BadWordError("The file name has a bad word.")
    filename = os.path.join(TEMP_DIR, MODEL_PREFIX + str(random.randint(1, 1000)) + ".zip")
    response = requests.get(url)
    total = int(response.headers.get('content-length', 0)) 
    if total > 500000000:

        return ["The file is too large. You can only download files up to 500 MB in size.", "Failed", "Failed"]
    current = 0
    with open(filename, "wb") as f:
        for data in response.iter_content(chunk_size=4096): 
            f.write(data)
            current += len(data)
            print(progress_bar(total, current), end="\r") #
    
    

    try:
        unzip_file(filename)
    except Exception as e:
        return ["Failed to unzip the file", "Failed", "Failed"] 
    unzipped_dir = os.path.join(TEMP_DIR, os.path.basename(filename).split(".")[0])
    pth_files = []
    index_files = []
    for root, dirs, files in os.walk(unzipped_dir): 
        for file in files:
            if file.endswith(".pth"):
                pth_files.append(os.path.join(root, file))
            elif file.endswith(".index"):
                index_files.append(os.path.join(root, file))
    
    print(pth_files, index_files) 
    global pth_file
    global index_file
    pth_file = pth_files[0]
    index_file = index_files[0]

    print(pth_file)
    print(index_file)

    if name == "":
        name = pth_file.split(".")[0]

    MODELS.append({"model": pth_file, "index": index_file, "model_name": name})
    return ["Downloaded as " + name, pth_files[0], index_files[0]]

def inference(audio, model_name):
        output_data = inf_handler(audio, model_name)
        vocals = output_data[0]
        inst = output_data[1]

        return vocals, inst

if spaces_status:
    @spaces.GPU()
    def convert_now(audio_files, random_tag, converter):
        return converter(
            audio_files,
            random_tag,
            overwrite=False,
            parallel_workers=8
        )

        
else:
    def convert_now(audio_files, random_tag, converter):
        return converter(
            audio_files,
            random_tag,
            overwrite=False,
            parallel_workers=8
        )

def calculate_remaining_time(epochs, seconds_per_epoch):
    total_seconds = epochs * seconds_per_epoch

    hours = total_seconds // 3600
    minutes = (total_seconds % 3600) // 60
    seconds = total_seconds % 60

    if hours == 0:
        return f"{int(minutes)} minutes"
    elif hours == 1:
        return f"{int(hours)} hour and {int(minutes)} minutes"
    else:
        return f"{int(hours)} hours and {int(minutes)} minutes"

def inf_handler(audio, model_name): 
    model_found = False
    for model_info in UVR_5_MODELS:
        if model_info["model_name"] == model_name:
            separator.load_model(model_info["checkpoint"])
            model_found = True
            break
    if not model_found:
        separator.load_model()
    output_files = separator.separate(audio)
    vocals = output_files[0]
    inst = output_files[1]
    return vocals, inst

    
def run(
    model,
    audio_files,
    pitch_alg,
    pitch_lvl,
    index_inf,
    r_m_f,
    e_r,
    c_b_p,
):
    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)

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

    file_m = model
    print("File model:", file_m)

    # get from MODELS
    for model in MODELS:
        if model["model_name"] == file_m:
            print(model)
            file_m = model["model"]
            file_index = model["index"]
            break
    
    if not file_m.endswith(".pth"):
        raise ValueError("The model file must be a .pth file")


    print("Random tag:", random_tag)
    print("File model:", file_m)
    print("Pitch algorithm:", pitch_alg)
    print("Pitch level:", pitch_lvl)
    print("File index:", file_index)
    print("Index influence:", index_inf)
    print("Respiration median filtering:", r_m_f)
    print("Envelope ratio:", e_r)

    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=44100 if audio_files[0].endswith('.mp3') else 0, 
    )
    time.sleep(0.1)

    result = convert_now(audio_files, random_tag, converter)
    print("Result:", result)

    return result[0]

def upload_model(index_file, pth_file, model_name):
    pth_file = pth_file.name
    index_file = index_file.name
    MODELS.append({"model": pth_file, "index": index_file, "model_name": model_name})
    return "Uploaded!"  

with gr.Blocks(theme=gr.themes.Default(primary_hue="pink", secondary_hue="rose"), title="Ilaria RVC 💖") as demo:
    gr.Markdown("## Ilaria RVC 💖")
    with gr.Tab("Inference"):
        sound_gui = gr.Audio(value=None,type="filepath",autoplay=False,visible=True,)
        def update():
            print(MODELS)
            return gr.Dropdown(label="Model",choices=[model["model_name"] for model in MODELS],visible=True,interactive=True, value=MODELS[0]["model_name"],)
        with gr.Row():
            models_dropdown = gr.Dropdown(label="Model",choices=[model["model_name"] for model in MODELS],visible=True,interactive=True, value=MODELS[0]["model_name"],)
            refresh_button = gr.Button("Refresh Models")
            refresh_button.click(update, outputs=[models_dropdown])

        with gr.Accordion("Ilaria TTS", open=False):
            text_tts = gr.Textbox(label="Text", placeholder="Hello!", lines=3, interactive=True,)
            dropdown_tts = gr.Dropdown(label="Language and Model",choices=list(language_dict.keys()),interactive=True, value=list(language_dict.keys())[0])

            button_tts = gr.Button("Speak", variant="primary",)
            button_tts.click(text_to_speech_edge, inputs=[text_tts, dropdown_tts], outputs=[sound_gui])

        with gr.Accordion("Settings", open=False):
            pitch_algo_conf = gr.Dropdown(PITCH_ALGO_OPT,value=PITCH_ALGO_OPT[4],label="Pitch algorithm",visible=True,interactive=True,)
            pitch_lvl_conf = gr.Slider(label="Pitch level (lower -> 'male' while higher -> 'female')",minimum=-24,maximum=24,step=1,value=0,visible=True,interactive=True,)
            index_inf_conf =  gr.Slider(minimum=0,maximum=1,label="Index influence -> How much accent is applied",value=0.75,)
            respiration_filter_conf = gr.Slider(minimum=0,maximum=7,label="Respiration median filtering",value=3,step=1,interactive=True,)
            envelope_ratio_conf = gr.Slider(minimum=0,maximum=1,label="Envelope ratio",value=0.25,interactive=True,)
            consonant_protec_conf = gr.Slider(minimum=0,maximum=0.5,label="Consonant breath protection",value=0.5,interactive=True,)

        button_conf = gr.Button("Convert",variant="primary",)
        output_conf = gr.Audio(type="filepath",label="Output",)
    	
        button_conf.click(lambda :None, None, output_conf)
        button_conf.click(
            run,
            inputs=[
                models_dropdown,
                sound_gui,
                pitch_algo_conf,
                pitch_lvl_conf,
                index_inf_conf,
                respiration_filter_conf,
                envelope_ratio_conf,
                consonant_protec_conf,
            ],
            outputs=[output_conf],
        )


    with gr.Tab("Model Loader (Download and Upload)"):
        with gr.Accordion("Model Downloader", open=False):
            gr.Markdown(
                "Download the model from the following URL and upload it here. (Huggingface RVC model)"
            )
            model = gr.Textbox(lines=1, label="Model URL")
            name = gr.Textbox(lines=1, label="Model Name", placeholder="Model Name")
            download_button = gr.Button("Download Model")
            status = gr.Textbox(lines=1, label="Status", placeholder="Waiting....", interactive=False)
            model_pth = gr.Textbox(lines=1, label="Model pth file", placeholder="Waiting....", interactive=False)
            index_pth = gr.Textbox(lines=1, label="Index pth file", placeholder="Waiting....", interactive=False)
            download_button.click(download_from_url, [model, name], outputs=[status, model_pth, index_pth])
        with gr.Accordion("Upload A Model", open=False):
            index_file_upload = gr.File(label="Index File (.index)")
            pth_file_upload = gr.File(label="Model File (.pth)")

            model_name = gr.Textbox(label="Model Name", placeholder="Model Name")
            upload_button = gr.Button("Upload Model")
            upload_status = gr.Textbox(lines=1, label="Status", placeholder="Waiting....", interactive=False)

            upload_button.click(upload_model, [index_file_upload, pth_file_upload, model_name], upload_status)
    

    with gr.Tab("Vocal Separator (UVR)"):
        gr.Markdown("Separate vocals and instruments from an audio file using UVR models. - This is only on CPU due to ZeroGPU being ZeroGPU :(")
        uvr5_audio_file = gr.Audio(label="Audio File",type="filepath")

        with gr.Row():
            uvr5_model = gr.Dropdown(label="Model", choices=[model["model_name"] for model in UVR_5_MODELS])
            uvr5_button = gr.Button("Separate Vocals", variant="primary",)

        uvr5_output_voc = gr.Audio(type="filepath", label="Output 1",)
        uvr5_output_inst = gr.Audio(type="filepath", label="Output 2",)

        uvr5_button.click(inference, [uvr5_audio_file, uvr5_model], [uvr5_output_voc, uvr5_output_inst])
    
    with gr.Tab("Extra"):
        with gr.Accordion("Model Information", open=False):
            def json_to_markdown_table(json_data):
                table = "| Key | Value |\n| --- | --- |\n"
                for key, value in json_data.items():
                    table += f"| {key} | {value} |\n"
                return table
            def model_info(name):
                for model in MODELS:
                    if model["model_name"] == name:
                        print(model["model"])
                        info = model_handler.model_info(model["model"])
                        info2 = {
                            "Model Name": model["model_name"],
                            "Model Config": info['config'],
                            "Epochs Trained": info['epochs'],
                            "Sample Rate": info['sr'],
                            "Pitch Guidance": info['f0'],
                            "Model Precision": info['size'],
                        }
                        return gr.Markdown(json_to_markdown_table(info2))

                return "Model not found"
            def update():
                print(MODELS)
                return gr.Dropdown(label="Model", choices=[model["model_name"] for model in MODELS])
            with gr.Row():
                model_info_dropdown = gr.Dropdown(label="Model", choices=[model["model_name"] for model in MODELS])
                refresh_button = gr.Button("Refresh Models")
                refresh_button.click(update, outputs=[model_info_dropdown])
            model_info_button = gr.Button("Get Model Information")
            model_info_output = gr.Textbox(value="Waiting...",label="Output", interactive=False)
            model_info_button.click(model_info, [model_info_dropdown], [model_info_output])
            


        with gr.Accordion("Training Time Calculator", open=False):
            with gr.Column():
                epochs_input = gr.Number(label="Number of Epochs")
                seconds_input = gr.Number(label="Seconds per Epoch")
                calculate_button = gr.Button("Calculate Time Remaining")
                remaining_time_output = gr.Textbox(label="Remaining Time", interactive=False)
                
                calculate_button.click(calculate_remaining_time,inputs=[epochs_input, seconds_input],outputs=[remaining_time_output])

        with gr.Accordion("Model Fusion", open=False): 
                    with gr.Group():
                        def merge(ckpt_a, ckpt_b, alpha_a, sr_, if_f0_, info__, name_to_save0, version_2):
                            for model in MODELS:
                                if model["model_name"] == ckpt_a:
                                    ckpt_a = model["model"]
                                if model["model_name"] == ckpt_b:
                                    ckpt_b = model["model"]
                            
                            path = model_handler.merge(ckpt_a, ckpt_b, alpha_a, sr_, if_f0_, info__, name_to_save0, version_2)
                            if path == "Fail to merge the models. The model architectures are not the same.":
                                return "Fail to merge the models. The model architectures are not the same."
                            else:
                                MODELS.append({"model": path, "index": None, "model_name": name_to_save0})
                                return "Merged, saved as " + name_to_save0

                        gr.Markdown(value="Strongly suggested to use only very clean models.")
                        with gr.Row():
                            def update():
                                print(MODELS)
                                return gr.Dropdown(label="Model A", choices=[model["model_name"] for model in MODELS]), gr.Dropdown(label="Model B", choices=[model["model_name"] for model in MODELS])
                            refresh_button_fusion = gr.Button("Refresh Models")
                            ckpt_a = gr.Dropdown(label="Model A", choices=[model["model_name"] for model in MODELS])
                            ckpt_b = gr.Dropdown(label="Model B", choices=[model["model_name"] for model in MODELS])
                            refresh_button_fusion.click(update, outputs=[ckpt_a, ckpt_b])
                            alpha_a = gr.Slider(
                                minimum=0,
                                maximum=1,
                                label="Weight of the first model over the second",
                                value=0.5,
                                interactive=True,
                            )
                    with gr.Group():
                        with gr.Row():
                            sr_ = gr.Radio(
                                label="Sample rate of both models",
                                choices=["32k","40k", "48k"],
                                value="32k",
                                interactive=True,
                            )
                            if_f0_ = gr.Radio(
                                label="Pitch Guidance",
                                choices=["Yes", "Nah"],
                                value="Yes",
                                interactive=True,
                            )
                            info__ = gr.Textbox(
                                label="Add informations to the model",
                                value="",
                                max_lines=8,
                                interactive=True,
                                visible=False
                            )
                            name_to_save0 = gr.Textbox(
                                label="Final Model name",
                                value="",
                                max_lines=1,
                                interactive=True,
                            )
                            version_2 = gr.Radio(
                                label="Versions of the models",
                                choices=["v1", "v2"],
                                value="v2",
                                interactive=True,
                            )
                    with gr.Group():
                        with gr.Row():
                            but6 = gr.Button("Fuse the two models", variant="primary")
                            info4 = gr.Textbox(label="Output", value="", max_lines=8)
                        but6.click(
                            merge,
                            [ckpt_a,ckpt_b,alpha_a,sr_,if_f0_,info__,name_to_save0,version_2,],info4,api_name="ckpt_merge",)

        with gr.Accordion("Model Quantization", open=False):
            gr.Markdown("Quantize the model to a lower precision. - soon™ or never™ 😎")

        with gr.Accordion("Debug", open=False):
            def json_to_markdown_table(json_data):
                table = "| Key | Value |\n| --- | --- |\n"
                for key, value in json_data.items():
                    table += f"| {key} | {value} |\n"
                return table
            gr.Markdown("View the models that are currently loaded in the instance.")

            gr.Markdown(json_to_markdown_table({"Models": len(MODELS), "UVR Models": len(UVR_5_MODELS)}))

            gr.Markdown("View the current status of the instance.")
            status = {
                "Status": "Running", # duh lol
                "Models": len(MODELS),
                "UVR Models": len(UVR_5_MODELS),
                "CPU Usage": f"{psutil.cpu_percent()}%",
                "RAM Usage": f"{psutil.virtual_memory().percent}%",
                "CPU": f"{cpuinfo.get_cpu_info()['brand_raw']}",
                "System Uptime": f"{round(time.time() - psutil.boot_time(), 2)} seconds",
                "System Load Average": f"{psutil.getloadavg()}",
                "====================": "====================",
                "CPU Cores": psutil.cpu_count(),
                "CPU Threads": psutil.cpu_count(logical=True),
                "RAM Total": f"{round(psutil.virtual_memory().total / 1024**3, 2)} GB",
                "RAM Used": f"{round(psutil.virtual_memory().used / 1024**3, 2)} GB",
                "CPU Frequency": f"{psutil.cpu_freq().current} MHz",
                "====================": "====================",
                "GPU": "A100 - Do a request (Inference, you won't see it either way)",
            }
            gr.Markdown(json_to_markdown_table(status))

    with gr.Tab("Credits"):
        gr.Markdown(
            """
            Ilaria RVC made by [Ilaria](https://huggingface.co/TheStinger) suport her on [ko-fi](https://ko-fi.com/ilariaowo)
            
            The Inference code is made by [r3gm](https://huggingface.co/r3gm) (his module helped form this space 💖)

            made with ❤️ by [mikus](https://github.com/cappuch) - made the ui!

            ## In loving memory of JLabDX 🕊️
            """
        )
    with gr.Tab(("")):
        gr.Markdown('''
            ![ilaria](https://i.ytimg.com/vi/5PWqt2Wg-us/maxresdefault.jpg)
        ''')

demo.queue(api_open=False).launch(show_api=False)