app.py

#
# SPDX-FileCopyrightText: Hadad <hadad@linuxmail.org>
# SPDX-License-Identifier: Apache-2.0
#

import math
import torch
import gradio as gr
torch.set_num_threads(1)
torch.set_num_interop_threads(1)
from config import (
    AVAILABLE_VOICES,
    DEFAULT_VOICE,
    DEFAULT_MODEL_VARIANT,
    DEFAULT_TEMPERATURE,
    DEFAULT_LSD_DECODE_STEPS,
    DEFAULT_EOS_THRESHOLD,
    DEFAULT_NOISE_CLAMP,
    DEFAULT_FRAMES_AFTER_EOS,
    MAXIMUM_INPUT_LENGTH,
    VOICE_MODE_PRESET,
    VOICE_MODE_CLONE,
    EXAMPLE_PROMPTS
)
from src.core.authentication import authenticate_huggingface
authenticate_huggingface()
from src.core.memory import start_background_cleanup_thread
start_background_cleanup_thread()
from src.generation.handler import (
    perform_speech_generation,
    request_generation_stop
)
from src.ui.state import (
    check_generate_button_state,
    calculate_character_count_display,
    determine_clear_button_visibility,
    update_voice_mode_visibility
)
from src.ui.handlers import (
    switch_to_generating_state,
    switch_to_idle_state,
    perform_clear_action,
    create_example_handler,
    format_example_button_label
)
from assets.css.styles import CSS
from assets.static.title import TITLE
from assets.static.header import HEADER
from assets.static.footer import FOOTER
from assets.static.sidebar import SIDEBAR

with gr.Blocks(css=CSS, fill_height=False, fill_width=True) as app:
    ui_state = gr.State({"generating": False})

    with gr.Sidebar():
        gr.HTML(SIDEBAR())

    with gr.Column(elem_classes="header-section"):
        gr.HTML(TITLE())
        gr.HTML(HEADER())

    with gr.Row():
        with gr.Column():
            audio_output_component = gr.Audio(
                label="Generated Speech Output",
                type="filepath",
                interactive=False,
                autoplay=False
            )

            with gr.Accordion("Voice Selection", open=True):
                voice_mode_radio = gr.Radio(
                    label="Voice Mode",
                    choices=[
                        VOICE_MODE_PRESET,
                        VOICE_MODE_CLONE
                    ],
                    value=VOICE_MODE_PRESET,
                    info="Choose between preset voices or clone a voice from uploaded audio",
                    elem_id="voice-mode"
                )

                with gr.Column(visible=True) as preset_voice_container:
                    voice_preset_dropdown = gr.Dropdown(
                        label="Select Preset Voice",
                        choices=AVAILABLE_VOICES,
                        value=DEFAULT_VOICE
                    )

                with gr.Column(visible=False) as clone_voice_container:
                    voice_clone_audio_input = gr.Audio(
                        label="Upload Audio for Voice Cloning",
                        type="filepath"
                    )

            with gr.Accordion("Model Parameters", open=False):
                with gr.Row():
                    temperature_slider = gr.Slider(
                        label="Temperature",
                        minimum=0.1,
                        maximum=2.0,
                        step=0.05,
                        value=DEFAULT_TEMPERATURE,
                        info="Higher values produce more expressive speech"
                    )
                    
                    lsd_decode_steps_slider = gr.Slider(
                        label="LSD Decode Steps",
                        minimum=1,
                        maximum=20,
                        step=1,
                        value=DEFAULT_LSD_DECODE_STEPS,
                        info="More steps may improve quality but slower"
                    )

                with gr.Row():
                    noise_clamp_slider = gr.Slider(
                        label="Noise Clamp",
                        minimum=0.0,
                        maximum=2.0,
                        step=0.05,
                        value=DEFAULT_NOISE_CLAMP,
                        info="Maximum noise sampling value (0 = disabled)"
                    )
                    
                    eos_threshold_slider = gr.Slider(
                        label="End of Sequence Threshold",
                        minimum=-10.0,
                        maximum=0.0,
                        step=0.25,
                        value=DEFAULT_EOS_THRESHOLD,
                        info="Smaller values cause earlier completion"
                    )

            with gr.Accordion("Advanced Settings", open=False):
                model_variant_textbox = gr.Textbox(
                    label="Model Variant Identifier",
                    value=DEFAULT_MODEL_VARIANT,
                    info="Model signature for generation"
                )

                with gr.Row():
                    enable_custom_frames_checkbox = gr.Checkbox(
                        label="Enable Custom Frames After EOS",
                        value=False,
                        info="Manually control post-EOS frame generation"
                    )
                    
                    frames_after_eos_slider = gr.Slider(
                        label="Frames After EOS",
                        minimum=0,
                        maximum=100,
                        step=1,
                        value=DEFAULT_FRAMES_AFTER_EOS,
                        info="Additional frames after end-of-sequence (80ms per frame)"
                    )

        with gr.Column(scale=1):
            text_input_component = gr.Textbox(
                label="Prompt",
                placeholder="Enter the text you want to convert to speech...",
                lines=2,
                max_lines=20,
                max_length=MAXIMUM_INPUT_LENGTH,
                autoscroll=True
            )

            character_count_display = gr.HTML(
                f"<div style='text-align: right; padding: 4px 0;'><span style='color: var(--body-text-color-subdued); font-size: 0.85em;'>0 / {MAXIMUM_INPUT_LENGTH}</span></div>",
                visible=False
            )

            generate_button = gr.Button(
                "Generate",
                variant="primary",
                size="lg",
                interactive=False
            )

            stop_button = gr.Button(
                "Stop",
                variant="stop",
                size="lg",
                visible=False
            )

            clear_button = gr.Button(
                "Clear",
                variant="secondary",
                size="lg",
                visible=False
            )

            gr.HTML(
                """
                <div style="padding: 16px 0 8px 0;">
                    <h3 style="margin: 0 0 8px 0; font-size: 1.1em;">Example Prompts</h3>
                    <p style="margin: 0; opacity: 0.7; font-size: 0.9em;">Click any example to generate speech with its assigned voice</p>
                </div>
                """
            )

            example_buttons_list = []
            num_examples = len(EXAMPLE_PROMPTS)
            examples_per_row = 2
            num_rows = math.ceil(num_examples / examples_per_row)

            for row_idx in range(num_rows):
                with gr.Row():
                    start_idx = row_idx * examples_per_row
                    end_idx = min(start_idx + examples_per_row, num_examples)
                    for i in range(start_idx, end_idx):
                        btn = gr.Button(
                            format_example_button_label(
                                EXAMPLE_PROMPTS[i]["text"],
                                EXAMPLE_PROMPTS[i]["voice"]
                            ),
                            size="sm",
                            variant="secondary"
                        )
                        example_buttons_list.append(btn)

    gr.HTML(FOOTER())

    generation_inputs = [
        text_input_component,
        voice_mode_radio,
        voice_preset_dropdown,
        voice_clone_audio_input,
        model_variant_textbox,
        lsd_decode_steps_slider,
        temperature_slider,
        noise_clamp_slider,
        eos_threshold_slider,
        frames_after_eos_slider,
        enable_custom_frames_checkbox
    ]

    voice_mode_radio.change(
        fn=update_voice_mode_visibility,
        inputs=[voice_mode_radio],
        outputs=[
            preset_voice_container,
            clone_voice_container
        ]
    )

    text_input_component.change(
        fn=calculate_character_count_display,
        inputs=[text_input_component],
        outputs=[character_count_display]
    )

    text_input_component.change(
        fn=check_generate_button_state,
        inputs=[
            text_input_component,
            ui_state
        ],
        outputs=[generate_button]
    )

    text_input_component.change(
        fn=determine_clear_button_visibility,
        inputs=[
            text_input_component,
            ui_state
        ],
        outputs=[clear_button]
    )

    generate_button.click(
        fn=switch_to_generating_state,
        inputs=[ui_state],
        outputs=[
            generate_button,
            stop_button,
            clear_button,
            ui_state
        ]
    ).then(
        fn=perform_speech_generation,
        inputs=generation_inputs,
        outputs=[audio_output_component]
    ).then(
        fn=switch_to_idle_state,
        inputs=[
            text_input_component,
            ui_state
        ],
        outputs=[
            generate_button,
            stop_button,
            clear_button,
            ui_state
        ]
    )

    stop_button.click(
        fn=request_generation_stop,
        outputs=[stop_button]
    )

    clear_button.click(
        fn=perform_clear_action,
        outputs=[
            text_input_component,
            audio_output_component,
            clear_button,
            voice_mode_radio,
            voice_preset_dropdown,
            voice_clone_audio_input
        ]
    )

    for button_index, example_button in enumerate(example_buttons_list):
        example_text = EXAMPLE_PROMPTS[button_index]["text"]
        example_voice = EXAMPLE_PROMPTS[button_index]["voice"]

        example_button.click(
            fn=switch_to_generating_state,
            inputs=[ui_state],
            outputs=[
                generate_button,
                stop_button,
                clear_button,
                ui_state
            ]
        ).then(
            fn=create_example_handler(example_text, example_voice),
            outputs=[
                text_input_component,
                voice_mode_radio,
                voice_preset_dropdown
            ]
        ).then(
            fn=perform_speech_generation,
            inputs=generation_inputs,
            outputs=[audio_output_component]
        ).then(
            fn=switch_to_idle_state,
            inputs=[
                text_input_component,
                ui_state
            ],
            outputs=[
                generate_button,
                stop_button,
                clear_button,
                ui_state
            ]
        )

app.launch(server_name="0.0.0.0")