app.py

import re
from datetime import datetime
from typing import Tuple

import gradio as gr
import numpy as np
import torch
from espnet2.bin.tts_inference import Text2Speech

model = Text2Speech.from_pretrained("yesyesjaewook/jets-jaewook-ko")

emotions = {
    "가엾다": 1,
    "갑갑하다": 2,
    "갑작스럽다": 3,
    "경멸스럽다": 4,
    "고맙다": 5,
    "고통스럽다": 6,
    "곤란하다": 7,
    "괴롭다": 8,
    "굉장하다": 9,
    "그립다": 10,
    "급하다": 11,
    "기막히다": 12,
    "긴밀하다": 13,
    "꺼림칙하다": 14,
    "끔찍하다": 15,
    "난데없다": 16,
    "남사스럽다": 17,
    "냉정하다": 18,
    "녹녹하다": 19,
    "다급하다": 20,
    "담백하다": 21,
    "대동소이하다": 22,
    "두렵다": 23,
    "둥그스름하다": 24,
    "뚜렷하다": 25,
    "만족스럽다": 26,
    "망하다": 27,
    "매정하다": 28,
    "못되다": 29,
    "무관심하다": 30,
    "무례하다": 31,
    "뭉클하다": 32,
    "밉다": 33,
    "반갑다": 34,
    "배은망덕하다": 35,
    "버겁다": 36,
    "보잘것없다": 37,
    "부끄럽다": 38,
    "부당하다": 39,
    "부유하다": 40,
    "불쌍하다": 41,
    "불행하다": 42,
    "불확실하다": 43,
    "뻔뻔스럽다": 44,
    "뼈아프다": 45,
    "사랑스럽다": 46,
    "서럽다": 47,
    "섭섭하다": 48,
    "수상하다": 49,
    "슬프다": 50,
    "시무룩하다": 51,
    "심란하다": 52,
    "쓰리다": 53,
    "쓸쓸하다": 54,
    "아니꼽다": 55,
    "아득하다": 56,
    "아쉽다": 57,
    "아찔하다": 58,
    "악랄하다": 59,
    "안타깝다": 60,
    "약하다": 61,
    "얕다": 62,
    "어렴풋하다": 63,
    "어리둥절하다": 64,
    "억울하다": 65,
    "언짢다": 66,
    "엄밀하다": 67,
    "엄중하다": 68,
    "엉뚱하다": 69,
    "영광스럽다": 70,
    "예사롭다": 71,
    "외롭다": 72,
    "위험하다": 73,
    "의심스럽다": 74,
    "자랑스럽다": 75,
    "자세하다": 76,
    "자유롭다": 77,
    "조마조마하다": 78,
    "즐겁다": 79,
    "짜증스럽다": 80,
    "창피하다": 81,
    "치욕스럽다": 82,
    "한심하다": 83,
    "행복하다": 84,
    "혼란하다": 85,
    "홀가분하다": 86,
    "화목하다": 87,
    "흐뭇하다": 88,
}


def float32_to_pcm16(waveform: torch.Tensor) -> np.array:
    info = np.iinfo(np.int16)
    waveform = waveform.numpy()
    return (waveform * info.max).clip(info.min, info.max).astype(np.int16)


def endswith_punctuation(text: str) -> bool:
    return re.search(r"[.?!]", text) is not None


def synthesize(text: str, emotion: str) -> Tuple[int, np.array]:
    text = text.strip()
    if not endswith_punctuation(text):
        text += "."

    print(f"[{datetime.now().isoformat()}] <{emotion}> {text}")

    output = model(text, lids=np.array(emotions[emotion]))
    return (model.fs, float32_to_pcm16(output["wav"]))


with gr.Blocks() as demo:
    gr.Markdown("# 송재욱 TTS 프로젝트")

    with gr.Row():
        with gr.Column():
            text = gr.Textbox(
                label="텍스트",
                value="그동안 사랑해주신분들에게는 감사하다는 말만 전하고싶습니다. 제 방송 햇수로 3년이 흐르고 지금 여기까지 오는데 굉장히 노력많이했습니다.",
                lines=3,
            )
            synthesize_button = gr.Button("합성")
            with gr.Box():
                ouptut = gr.Audio()
                gr.Markdown("음성 다운로드는 플레이어 오른쪽의 ··· 메뉴 클릭 부탁드립니다 🙏")

        with gr.Column():
            emotion = gr.Radio(label="감정", choices=[*emotions.keys()], value="담백하다")

    synthesize_button.click(fn=synthesize, inputs=[text, emotion], outputs=ouptut)

demo.launch()