app.py

import subprocess

# Define the local paths to the packages
local_package_paths = ["./TTS", "./transformers"]

# Run the pip install command for each local package
for package_path in local_package_paths:
    subprocess.run(["pip", "install", "-e", package_path])

import gradio as gr
from share_btn import community_icon_html, loading_icon_html, share_js
import os 
import shutil
import re

#from huggingface_hub import snapshot_download
import numpy as np
from scipy.io import wavfile
from scipy.io.wavfile import write, read
from pydub import AudioSegment

file_upload_available = os.environ.get("ALLOW_FILE_UPLOAD")
MAX_NUMBER_SENTENCES = 10

import json
with open("characters.json", "r") as file:
    data = json.load(file)
    characters = [
        {
            "image": item["image"],
            "title": item["title"],
            "speaker": item["speaker"]
        }
        for item in data
    ]
    
from TTS.api import TTS
tts = TTS("tts_models/multilingual/multi-dataset/bark", gpu=True)

def cut_wav(input_path, max_duration):
    # Load the WAV file
    audio = AudioSegment.from_wav(input_path)
    
    # Calculate the duration of the audio
    audio_duration = len(audio) / 1000  # Convert milliseconds to seconds
    
    # Determine the duration to cut (maximum of max_duration and actual audio duration)
    cut_duration = min(max_duration, audio_duration)
    
    # Cut the audio
    cut_audio = audio[:int(cut_duration * 1000)]  # Convert seconds to milliseconds
    
    # Get the input file name without extension
    file_name = os.path.splitext(os.path.basename(input_path))[0]
    
    # Construct the output file path with the original file name and "_cut" suffix
    output_path = f"{file_name}_cut.wav"
    
    # Save the cut audio as a new WAV file
    cut_audio.export(output_path, format="wav")

    return output_path

def load_hidden(audio_in):
    return audio_in

def load_hidden_mic(audio_in):
    print("USER RECORDED A NEW SAMPLE")
    
    library_path = 'bark_voices'  
    folder_name = 'audio-0-100'  
    second_folder_name = 'audio-0-100_cleaned' 
    
    folder_path = os.path.join(library_path, folder_name)
    second_folder_path = os.path.join(library_path, second_folder_name)

    print("We need to clean previous util files, if needed:")
    if os.path.exists(folder_path):
        try:
            shutil.rmtree(folder_path)
            print(f"Successfully deleted the folder previously created from last raw recorded sample: {folder_path}")
        except OSError as e:
            print(f"Error: {folder_path} - {e.strerror}")
    else:
        print(f"OK, the folder for a raw recorded sample does not exist: {folder_path}")

    if os.path.exists(second_folder_path):
        try:
            shutil.rmtree(second_folder_path)
            print(f"Successfully deleted the folder previously created from last cleaned recorded sample: {second_folder_path}")
        except OSError as e:
            print(f"Error: {second_folder_path} - {e.strerror}")
    else:
        print(f"Ok, the folder for a cleaned recorded sample does not exist: {second_folder_path}")
    
    return audio_in

def clear_clean_ckeck():
    return False

def wipe_npz_file(folder_path):
    print("YO • a user is manipulating audio inputs")
    
def split_process(audio, chosen_out_track):
    gr.Info("Cleaning your audio sample...")
    os.makedirs("out", exist_ok=True)
    write('test.wav', audio[0], audio[1])
    os.system("python3 -m demucs.separate -n mdx_extra_q -j 4 test.wav -o out")
    #return "./out/mdx_extra_q/test/vocals.wav","./out/mdx_extra_q/test/bass.wav","./out/mdx_extra_q/test/drums.wav","./out/mdx_extra_q/test/other.wav"
    if chosen_out_track == "vocals":
        print("Audio sample cleaned")
        return "./out/mdx_extra_q/test/vocals.wav"
    elif chosen_out_track == "bass":
        return "./out/mdx_extra_q/test/bass.wav"
    elif chosen_out_track == "drums":
        return "./out/mdx_extra_q/test/drums.wav"
    elif chosen_out_track == "other":
        return "./out/mdx_extra_q/test/other.wav"
    elif chosen_out_track == "all-in":
        return "test.wav"
        
def update_selection(selected_state: gr.SelectData):
    c_image = characters[selected_state.index]["image"]
    c_title = characters[selected_state.index]["title"]
    c_speaker = characters[selected_state.index]["speaker"]

    return c_title, selected_state

    
def infer(prompt, input_wav_file, clean_audio, hidden_numpy_audio):
    print("""
—————
NEW INFERENCE:
———————
    """)
    if prompt == "":
        gr.Warning("Do not forget to provide a tts prompt !")
    
    if clean_audio is True :
        print("We want to clean audio sample")
        # Extract the file name without the extension
        new_name = os.path.splitext(os.path.basename(input_wav_file))[0]
        print(f"FILE BASENAME is: {new_name}")
        if os.path.exists(os.path.join("bark_voices", f"{new_name}_cleaned")):
            print("This file has already been cleaned")
            check_name = os.path.join("bark_voices", f"{new_name}_cleaned")
            source_path = os.path.join(check_name, f"{new_name}_cleaned.wav")
        else:
            print("This file is new, we need to clean and store it")
            source_path = split_process(hidden_numpy_audio, "vocals")
        
            # Rename the file
            new_path = os.path.join(os.path.dirname(source_path), f"{new_name}_cleaned.wav")
            os.rename(source_path, new_path)
            source_path = new_path
    else :
        print("We do NOT want to clean audio sample")
        # Path to your WAV file
        source_path = input_wav_file

    # Destination directory
    destination_directory = "bark_voices"

    # Extract the file name without the extension
    file_name = os.path.splitext(os.path.basename(source_path))[0]

    # Construct the full destination directory path
    destination_path = os.path.join(destination_directory, file_name)

    # Create the new directory
    os.makedirs(destination_path, exist_ok=True)

    # Move the WAV file to the new directory
    shutil.move(source_path, os.path.join(destination_path, f"{file_name}.wav"))

    # —————
    
    # Split the text into sentences based on common punctuation marks
    sentences = re.split(r'(?<=[.!?])\s+', prompt)

    if len(sentences) > MAX_NUMBER_SENTENCES:
        gr.Info("Your text is too long. To keep this demo enjoyable for everyone, we only kept the first 10 sentences :) Duplicate this space and set MAX_NUMBER_SENTENCES for longer texts ;)")
        # Keep only the first MAX_NUMBER_SENTENCES sentences
        first_nb_sentences = sentences[:MAX_NUMBER_SENTENCES]
    
        # Join the selected sentences back into a single string
        limited_prompt = ' '.join(first_nb_sentences)
        prompt = limited_prompt

    else:
        prompt = prompt

    gr.Info("Generating audio from prompt")
    tts.tts_to_file(text=prompt,
                file_path="output.wav",
                voice_dir="bark_voices/",
                speaker=f"{file_name}")

    # List all the files and subdirectories in the given directory
    contents = os.listdir(f"bark_voices/{file_name}")

    # Print the contents
    for item in contents:
        print(item)  
    print("Preparing final waveform video ...")
    tts_video = gr.make_waveform(audio="output.wav")
    print(tts_video)
    print("FINISHED")
    return "output.wav", tts_video, gr.update(value=f"bark_voices/{file_name}/{contents[1]}", visible=True), gr.Group.update(visible=True), destination_path

def infer_from_c(prompt, c_name):
    print("""
—————
NEW INFERENCE:
———————
    """)
    if prompt == "":
        gr.Warning("Do not forget to provide a tts prompt !")
        print("Warning about prompt sent to user")
        
    print(f"USING VOICE LIBRARY: {c_name}")
    # Split the text into sentences based on common punctuation marks
    sentences = re.split(r'(?<=[.!?])\s+', prompt)
    
    if len(sentences) > MAX_NUMBER_SENTENCES:
        gr.Info("Your text is too long. To keep this demo enjoyable for everyone, we only kept the first 10 sentences :) Duplicate this space and set MAX_NUMBER_SENTENCES for longer texts ;)")    
        # Keep only the first MAX_NUMBER_SENTENCES sentences
        first_nb_sentences = sentences[:MAX_NUMBER_SENTENCES]
    
        # Join the selected sentences back into a single string
        limited_prompt = ' '.join(first_nb_sentences)
        prompt = limited_prompt

    else:
        prompt = prompt

    
    if c_name == "":
        gr.Warning("Voice character is not properly selected. Please ensure that the name of the chosen voice is specified in the Character Name input.")
        print("Warning about Voice Name sent to user")
    else:
        print(f"Generating audio from prompt with {c_name} ;)")
        
    tts.tts_to_file(text=prompt,
                file_path="output.wav",
                voice_dir="examples/library/",
                speaker=f"{c_name}")
    
    print("Preparing final waveform video ...")
    tts_video = gr.make_waveform(audio="output.wav")
    print(tts_video)
    print("FINISHED")
    return "output.wav", tts_video, gr.update(value=f"examples/library/{c_name}/{c_name}.npz", visible=True), gr.Group.update(visible=True)


css = """
#col-container {max-width: 780px; margin-left: auto; margin-right: auto;}
a {text-decoration-line: underline; font-weight: 600;}
.mic-wrap > button {
    width: 100%;
    height: 60px;
    font-size: 1.4em!important;
}
.record-icon.svelte-1thnwz {
    display: flex;
    position: relative;
    margin-right: var(--size-2);
    width: unset;
    height: unset;
}
span.record-icon > span.dot.svelte-1thnwz {
    width: 20px!important;
    height: 20px!important;
}
.animate-spin {
  animation: spin 1s linear infinite;
}
@keyframes spin {
  from {
      transform: rotate(0deg);
  }
  to {
      transform: rotate(360deg);
  }
}
#share-btn-container {
  display: flex; 
  padding-left: 0.5rem !important; 
  padding-right: 0.5rem !important; 
  background-color: #000000; 
  justify-content: center; 
  align-items: center; 
  border-radius: 9999px !important; 
  max-width: 15rem;
  height: 36px;
}
div#share-btn-container > div {
    flex-direction: row;
    background: black;
    align-items: center;
}
#share-btn-container:hover {
  background-color: #060606;
}
#share-btn {
  all: initial; 
  color: #ffffff;
  font-weight: 600; 
  cursor:pointer; 
  font-family: 'IBM Plex Sans', sans-serif; 
  margin-left: 0.5rem !important; 
  padding-top: 0.5rem !important; 
  padding-bottom: 0.5rem !important;
  right:0;
}
#share-btn * {
  all: unset;
}
#share-btn-container div:nth-child(-n+2){
  width: auto !important;
  min-height: 0px !important;
}
#share-btn-container .wrap {
  display: none !important;
}
#share-btn-container.hidden {
  display: none!important;
}
img[src*='#center'] { 
    display: block;
    margin: auto;
}
.footer {
        margin-bottom: 45px;
        margin-top: 10px;
        text-align: center;
        border-bottom: 1px solid #e5e5e5;
    }
    .footer>p {
        font-size: .8rem;
        display: inline-block;
        padding: 0 10px;
        transform: translateY(10px);
        background: white;
    }
    .dark .footer {
        border-color: #303030;
    }
    .dark .footer>p {
        background: #0b0f19;
    }
.disclaimer {
    text-align: left;
}
.disclaimer > p {
    font-size: .8rem;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        
        gr.Markdown("""
        <h1 style="text-align: center;">Voice Cloning Demo</h1>
        """)
        with gr.Row():
            with gr.Column():
                prompt = gr.Textbox(
                    label = "Text to speech prompt",
                    info = "One or two sentences at a time is better* (max: 10)",
                    placeholder = "Hello friend! How are you today?",
                    elem_id = "tts-prompt"
                )

            
            with gr.Column():
                audio_in = gr.Audio(
                    label="WAV voice to clone", 
                    type="filepath",
                    source="upload",
                    interactive = False
                )
                hidden_audio_numpy = gr.Audio(type="numpy", visible=False)
                submit_btn = gr.Button("Submit")
                
                with gr.Tab("Microphone"):
                    texts_samples = gr.Textbox(label = "Helpers", 
                                               info = "You can read out loud one of these sentences if you do not know what to record :)",
                                               value = """"Jazz, a quirky mix of groovy saxophones and wailing trumpets, echoes through the vibrant city streets."
———
"A majestic orchestra plays enchanting melodies, filling the air with harmony."
———
"The exquisite aroma of freshly baked bread wafts from a cozy bakery, enticing passersby."
———
"A thunderous roar shakes the ground as a massive jet takes off into the sky, leaving trails of white behind."
———
"Laughter erupts from a park where children play, their innocent voices rising like tinkling bells."
———
"Waves crash on the beach, and seagulls caw as they soar overhead, a symphony of nature's sounds."
———
"In the distance, a blacksmith hammers red-hot metal, the rhythmic clang punctuating the day."
———
"As evening falls, a soft hush blankets the world, crickets chirping in a soothing rhythm."
                                               """,
                                               interactive = False,
                                               lines = 5
                                              )
                    micro_in = gr.Audio(
                                label="Record voice to clone", 
                                type="filepath",
                                source="microphone",
                                interactive = True
                            )
                    clean_micro = gr.Checkbox(label="Clean sample ?", value=False)
                    micro_submit_btn = gr.Button("Submit")
                
                audio_in.upload(fn=load_hidden, inputs=[audio_in], outputs=[hidden_audio_numpy], queue=False)
                micro_in.stop_recording(fn=load_hidden_mic, inputs=[micro_in], outputs=[hidden_audio_numpy], queue=False)


            with gr.Column():
        
                cloned_out = gr.Audio(
                    label="Text to speech output",
                    visible = False
                )
        
                video_out = gr.Video(
                    label = "Waveform video",
                    elem_id = "voice-video-out"
                )
                
                npz_file = gr.File(
                    label = ".npz file",
                    visible = False
                )

                folder_path = gr.Textbox(visible=False)


        
        audio_in.change(fn=wipe_npz_file, inputs=[folder_path], queue=False)
        micro_in.clear(fn=wipe_npz_file, inputs=[folder_path], queue=False)
    submit_btn.click(
        fn = infer,
        inputs = [
            prompt,
            audio_in,
            hidden_audio_numpy
        ],
        outputs = [
            cloned_out, 
            video_out,
            npz_file,
            folder_path
        ]
    )

    micro_submit_btn.click(
        fn = infer,
        inputs = [
            prompt,
            micro_in,
            clean_micro,
            hidden_audio_numpy
        ],
        outputs = [
            cloned_out, 
            video_out,
            npz_file,
            folder_path
        ]
    )

demo.queue(api_open=False, max_size=10).launch()