app.py

import sys
import os
import re
import time
import math
import torch
import random
import spaces
# By using XTTS you agree to CPML license https://coqui.ai/cpml
os.environ["COQUI_TOS_AGREED"] = "1"

import gradio as gr
from TTS.api import TTS
from TTS.utils.manage import ModelManager

max_64_bit_int = 2**63 - 1
model_names = TTS().list_models()
print(model_names.__dict__)
print(model_names.__dir__())
model_name = "tts_models/multilingual/multi-dataset/xtts_v2"
m = model_name

# Automatic device detection
if torch.cuda.is_available():
    # cuda only
    device_type = "cuda"
    device_selection = "cuda:0"
    data_type = torch.float16
else:
    # no GPU or Amd
    device_type = "cpu"
    device_selection = "cpu"
    data_type = torch.float32

tts = TTS(model_name, gpu=torch.cuda.is_available())
tts.to(device_type)

def update_output(output_number):
    return [
        gr.update(visible = (2 <= output_number)),
        gr.update(visible = (3 <= output_number)),
        gr.update(visible = (4 <= output_number)),
        gr.update(visible = (5 <= output_number)),
        gr.update(visible = (6 <= output_number)),
        gr.update(visible = (7 <= output_number)),
        gr.update(visible = (8 <= output_number)),
        gr.update(visible = (9 <= output_number))
    ]

def predict0(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 0, generation_number, temperature, is_randomize_seed, seed, progress)

def predict1(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 1, generation_number, temperature, is_randomize_seed, seed, progress)

def predict2(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 2, generation_number, temperature, is_randomize_seed, seed, progress)

def predict3(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 3, generation_number, temperature, is_randomize_seed, seed, progress)

def predict4(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 4, generation_number, temperature, is_randomize_seed, seed, progress)

def predict5(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 5, generation_number, temperature, is_randomize_seed, seed, progress)

def predict6(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 6, generation_number, temperature, is_randomize_seed, seed, progress)

def predict7(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 7, generation_number, temperature, is_randomize_seed, seed, progress)

def predict8(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, generation_number, temperature, is_randomize_seed, seed, progress = gr.Progress()):
    return predict(prompt, language, gender, audio_file_pth, mic_file_path, use_mic, 8, generation_number, temperature, is_randomize_seed, seed, progress)

def predict(
    prompt,
    language,
    gender,
    audio_file_pth,
    mic_file_path,
    use_mic,
    i,
    generation_number,
    temperature,
    is_randomize_seed,
    seed,
    progress = gr.Progress()
):
    if generation_number <= i:
        return (
                None,
                None,
            )
    start = time.time()
    progress(0, desc = "Preparing data...")

    if len(prompt) < 2:
        gr.Warning("Please give a longer prompt text")
        return (
                None,
                None,
            )
    if 50000 < len(prompt):
        gr.Warning("Text length limited to 50,000 characters for this demo, please try shorter text")
        return (
            None,
            None,
        )

    if use_mic:
        if mic_file_path is None:
            gr.Warning("Please record your voice with Microphone, or uncheck Use Microphone to use reference audios")
            return (
                None,
                None,
            )
        else:
            speaker_wav = mic_file_path
    else:
        speaker_wav = audio_file_pth

    if speaker_wav is None:
        if gender == "male":
            speaker_wav = "./examples/male.mp3"
        else:
            speaker_wav = "./examples/female.wav"
        
    output_filename = f"{i + 1}_{re.sub('[^a-zA-Z0-9]', '_', language)}_{re.sub('[^a-zA-Z0-9]', '_', prompt)}"[:180] + ".wav"
    
    try:
        if language == "fr":
            if m.find("your") != -1:
                language = "fr-fr"
        if m.find("/fr/") != -1:
            language = None
    
        predict_on_gpu(i, generation_number, prompt, speaker_wav, language, output_filename, temperature, is_randomize_seed, seed, progress)
    except RuntimeError as e :
        if "device-assert" in str(e):
            # cannot do anything on cuda device side error, need to restart
            gr.Warning("Unhandled Exception encounter, please retry in a minute")
            print("Cuda device-assert Runtime encountered need restart")
            sys.exit("Exit due to cuda device-assert")
        else:
            raise e
        
    end = time.time()
    secondes = int(end - start)
    minutes = math.floor(secondes / 60)
    secondes = secondes - (minutes * 60)
    hours = math.floor(minutes / 60)
    minutes = minutes - (hours * 60)
    information = ("Start again to get a different result. " if is_randomize_seed else "") + "The sound has been generated in " + ((str(hours) + " h, ") if hours != 0 else "") + ((str(minutes) + " min, ") if hours != 0 or minutes != 0 else "") + str(secondes) + " sec."

    return (
        output_filename,
        information,
    )

@spaces.GPU(duration=60)
def predict_on_gpu(
    i,
    generation_number,
    prompt,
    speaker_wav,
    language,
    output_filename,
    temperature,
    is_randomize_seed,
    seed,
    progress
):
    progress((i + .5) / generation_number, desc = "Generating the audio #" + str(i + 1) + "...")
    if is_randomize_seed:
        seed = random.randint(0, max_64_bit_int)
        
    random.seed(seed)
    torch.manual_seed(seed)
    
    tts.tts_to_file(
        text = prompt,
        file_path = output_filename,
        speaker_wav = speaker_wav,
        language = language,
        temperature = temperature
    )

with gr.Blocks() as interface:
    gr.HTML(
        """
        <h1><center>XTTS</center></h1>
        <big><center>Generate long vocal from text in several languages following voice freely, without account, without watermark and download it</center></big>
        <br/>
<a href="https://huggingface.co/coqui/XTTS-v1">XTTS</a> is a Voice generation model that lets you clone voices into different languages by using just a quick 3-second audio clip. 
<br/>
XTTS is built on previous research, like Tortoise, with additional architectural innovations and training to make cross-language voice cloning and multilingual speech generation possible. 
<br/>
This is the same model that powers our creator application <a href="https://coqui.ai">Coqui Studio</a> as well as the <a href="https://docs.coqui.ai">Coqui API</a>. In production we apply modifications to make low-latency streaming possible.
<br/>
Leave a star on the Github <a href="https://github.com/coqui-ai/TTS">TTS</a>, where our open-source inference and training code lives.
<br/>
<p>To avoid the queue, you can duplicate this space on CPU, GPU or ZERO space GPU:
<br/>
<a href="https://huggingface.co/spaces/Fabrice-TIERCELIN/Multi-language_Text-to-Speech?duplicate=true">
<img style="margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a>
</p>
        """
    )
    with gr.Column():
        prompt = gr.Textbox(
            label = "Text Prompt",
            info = "One or two sentences at a time is better",
            value = "Hello, World! Here is an example of light voice cloning. Try to upload your best audio samples quality",
            elem_id = "prompt-id",
        )
        with gr.Group():
            language = gr.Dropdown(
                label="Language",
            info="Select an output language for the synthesised speech",
            choices=[
                    ["Arabic", "ar"],
                    ["Brazilian Portuguese", "pt"],
                    ["Mandarin Chinese", "zh-cn"],
                    ["Czech", "cs"],
                    ["Dutch", "nl"],
                    ["English", "en"],
                    ["French", "fr"],
                    ["German", "de"],
                    ["Italian", "it"],
                    ["Polish", "pl"],
                    ["Russian", "ru"],
                    ["Spanish", "es"],
                    ["Turkish", "tr"]
            ],
            max_choices=1,
            value="en",
            elem_id = "language-id",
            )
            gr.HTML("More languages <a href='https://huggingface.co/spaces/Brasd99/TTS-Voice-Cloner'>here</a>")
        gender = gr.Radio(
            ["female", "male"],
            label="Gender",
            info="Gender of the voice",
            elem_id = "gender-id",
        )
        audio_file_pth = gr.Audio(
            label="Reference Audio",
            #info="Click on the ✎ button to upload your own target speaker audio",
            type="filepath",
            value=None,
            elem_id = "audio-file-pth-id",
        )
        mic_file_path = gr.Audio(
            sources=["microphone"],
            type="filepath",
            #info="Use your microphone to record audio",
            label="Use Microphone for Reference",
            elem_id = "mic-file-path-id",
        )
        use_mic = gr.Checkbox(
            label = "Check to use Microphone as Reference",
                    value = False,
                    info = "Notice: Microphone input may not work properly under traffic",
                              elem_id = "use-mic-id",
                             )
        generation_number = gr.Slider(
                 minimum = 1,
                 maximum = 9,
                 step = 1,
                 value = 1,
                 label = "Generation number",
                 info = "How many audios to generate",
                 elem_id = "generation-number-id"
             )
        with gr.Accordion("Advanced options", open = False):
             temperature = gr.Slider(
                 minimum = 0,
                 maximum = 10,
                 step = .1,
                 value = .75,
                 label = "Temperature",
                 info = "Maybe useless",
                 elem_id = "temperature-id"
             )
             randomize_seed = gr.Checkbox(
                 label = "\U0001F3B2 Randomize seed",
                 value = True,
                 info = "If checked, result is always different",
                 elem_id = "randomize-seed-id"
             )
             seed = gr.Slider(
                 minimum = 0,
                 maximum = max_64_bit_int,
                 step = 1,
                 randomize = True,
                 label = "Seed",
                 elem_id = "seed-id"
             )

        submit = gr.Button(
            "🚀 Speak",
            variant = "primary",
            elem_id = "submit-id"
        )

        synthesised_audio_1 = gr.Audio(
            label="Synthesised Audio #1",
            autoplay = False,
            elem_id = "synthesised-audio-1-id"
        )

        synthesised_audio_2 = gr.Audio(
            label="Synthesised Audio #2",
            autoplay = False,
            elem_id = "synthesised-audio-2-id",
            visible = False
        )

        synthesised_audio_3 = gr.Audio(
            label="Synthesised Audio #3",
            autoplay = False,
            elem_id = "synthesised-audio-3-id",
            visible = False
        )

        synthesised_audio_4 = gr.Audio(
            label="Synthesised Audio #4",
            autoplay = False,
            elem_id = "synthesised-audio-4-id",
            visible = False
        )

        synthesised_audio_5 = gr.Audio(
            label="Synthesised Audio #5",
            autoplay = False,
            elem_id = "synthesised-audio-5-id",
            visible = False
        )

        synthesised_audio_6 = gr.Audio(
            label="Synthesised Audio #6",
            autoplay = False,
            elem_id = "synthesised-audio-6-id",
            visible = False
        )

        synthesised_audio_7 = gr.Audio(
            label="Synthesised Audio #7",
            autoplay = False,
            elem_id = "synthesised-audio-7-id",
            visible = False
        )

        synthesised_audio_8 = gr.Audio(
            label="Synthesised Audio #8",
            autoplay = False,
            elem_id = "synthesised-audio-8-id",
            visible = False
        )

        synthesised_audio_9 = gr.Audio(
            label="Synthesised Audio #9",
            autoplay = False,
            elem_id = "synthesised-audio-9-id",
            visible = False
        )
        information = gr.HTML()

    submit.click(fn = update_output, inputs = [
        generation_number
    ], outputs = [
        synthesised_audio_2,
        synthesised_audio_3,
        synthesised_audio_4,
        synthesised_audio_5,
        synthesised_audio_6,
        synthesised_audio_7,
        synthesised_audio_8,
        synthesised_audio_9
    ], queue = False, show_progress = False).success(predict0, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_1,
        information
    ], scroll_to_output = True).success(predict1, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_2,
        information
    ], scroll_to_output = True).success(predict2, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_3,
        information
    ], scroll_to_output = True).success(predict3, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_4,
        information
    ], scroll_to_output = True).success(predict4, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_5,
        information
    ], scroll_to_output = True).success(predict5, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_6,
        information
    ], scroll_to_output = True).success(predict6, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_7,
        information
    ], scroll_to_output = True).success(predict7, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_8,
        information
    ], scroll_to_output = True).success(predict8, inputs = [
        prompt,
        language,
        gender,
        audio_file_pth,
        mic_file_path,
        use_mic,
        generation_number,
        temperature,
        randomize_seed,
        seed
    ], outputs = [
        synthesised_audio_9,
        information
    ], scroll_to_output = True)

interface.queue(max_size = 5).launch(debug=True)