app.py

"""
SMARTok 실시간 다국어 데모 (완전 수정본)
────────────────────────────────────────────
• 탭1  🎙️ 오디오 번역            : 마이크/파일 → 번역 + TTS
• 탭2  📄 문서·이미지 번역        : PDF / 이미지(OCR) → 번역
• 탭3  ⏱️ 실시간 1언어 번역       : 마이크 → 1개 언어 실시간 자막
• 탭4  🌏 실시간 4개 언어 번역    : 마이크 → 영·중·태·러 동시 자막
────────────────────────────────────────────
필수 apt  : tesseract-ocr libtesseract-dev ocrmypdf ffmpeg
필수 pip  : gradio>=5.33 openai python-dotenv pdfplumber ocrmypdf pillow
"""

import gradio as gr
import openai, os, io, tempfile, mimetypes, json, uuid
from dotenv import load_dotenv
import pdfplumber, ocrmypdf
from PIL import Image

# ────────────── 0. 공통 초기화 ────────────────────────────────────
load_dotenv()
api_key = os.getenv("OPENAI_API_KEY")
if not api_key:
    raise RuntimeError("OPENAI_API_KEY를 .env 파일에 설정하세요!")
client = openai.OpenAI(api_key=api_key)

LANGUAGES = [
    "Korean", "English", "Japanese", "Chinese",
    "Thai", "Russian", "Vietnamese",
    "Spanish", "French"
]
LANG_CODE = {
    "Korean":"ko","English":"en","Japanese":"ja","Chinese":"zh",
    "Thai":"th","Russian":"ru","Vietnamese":"vi",
    "Spanish":"es","French":"fr"
}
VOICE = {l: ("nova" if l in ["Korean","Japanese","Chinese"] else "alloy")
         for l in LANGUAGES}
FOUR = ["English","Chinese","Thai","Russian"]
STREAM_SEC = 4   # Whisper 호출 주기

# ────────────── 1. 유틸 함수 ─────────────────────────────────────
def _safe(v):
    if v is None:
        return None
    return v["name"] if isinstance(v, dict) else v

def _gpt(text, src, tgt):
    rsp = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[
            {"role":"system",
             "content":f"Translate the following {src} text to {tgt}. "
                        "Return only the translation."},
            {"role":"user","content":text}
        ],
        temperature=0.3,max_tokens=4096
    )
    return rsp.choices[0].message.content.strip()

def _tts(text, lang):
    rsp = client.audio.speech.create(
        model="tts-1",
        voice=VOICE.get(lang,"alloy"),
        input=text[:4096]
    )
    tmp = tempfile.NamedTemporaryFile(delete=False,suffix=".mp3")
    tmp.write(rsp.content); tmp.close()
    return tmp.name

# ────────────── 2. 오디오(단건) 번역 ─────────────────────────────
def translate_audio(audio_in, src, tgt):
    p = _safe(audio_in)
    if not p or not os.path.exists(p):
        return "⚠️ 음성 파일 필요", "", None
    with open(p,"rb") as f:
        stt = client.audio.transcriptions.create(
            model="whisper-1", file=f, language=LANG_CODE.get(src))
    orig = stt.text.strip()
    if not orig:
        return "⚠️ 음성 인식 실패", "", None
    trans = _gpt(orig, src, tgt)
    return orig, trans, _tts(trans, tgt)

# ────────────── 3. 문서 / 이미지 번역 ────────────────────────────
def translate_doc(file_in, src, tgt):
    p = _safe(file_in)
    if not p or not os.path.exists(p):
        return "⚠️ PDF/이미지 업로드", ""
    ext = os.path.splitext(p)[1].lower()
    mime = mimetypes.guess_type(p)[0] or ""
    try:
        # PDF 그대로
        if ext==".pdf" or "pdf" in mime:
            with pdfplumber.open(p) as pdf:
                txt = "\n".join(pg.extract_text() or "" for pg in pdf.pages[:5])
        # 이미지 → OCR PDF
        else:
            img_pdf = tempfile.NamedTemporaryFile(delete=False,suffix=".pdf").name
            Image.open(p).save(img_pdf,"PDF")
            ocr_pdf = tempfile.NamedTemporaryFile(delete=False,suffix=".pdf").name
            ocrmypdf.ocr(img_pdf, ocr_pdf,
                         lang=LANG_CODE.get(src,"eng"),
                         deskew=True,optimize=0,progress_bar=False)
            with pdfplumber.open(ocr_pdf) as pdf:
                txt = "\n".join(pg.extract_text() or "" for pg in pdf.pages)
    except Exception as e:
        return f"❌ 추출 오류: {e}", ""
    txt = txt.strip()
    if not txt:
        return "⚠️ 텍스트 추출 실패", ""
    return txt, _gpt(txt, src, tgt)

# ────────────── 4. 실시간 1언어 ──────────────────────────────────
def stream_single(audio_path, src, tgt, state):
    state = state or {"orig":"", "trans":""}
    if not audio_path or not os.path.exists(audio_path):
        return state["orig"], state["trans"], state
    with open(audio_path,"rb") as f:
        stt = client.audio.transcriptions.create(
            model="whisper-1", file=f, language=LANG_CODE.get(src))
    full = stt.text.strip()
    new = full[len(state["orig"]):]
    if new:
        state["orig"] = full
        state["trans"] += " " + _gpt(new, src, tgt)
    return state["orig"], state["trans"].strip(), state

# ────────────── 5. 실시간 4언어 ──────────────────────────────────
def stream_multi(audio_path, src, state):
    state = state or {k:"" for k in ["orig"]+FOUR}
    if not audio_path or not os.path.exists(audio_path):
        return state["orig"],state["English"],state["Chinese"],state["Thai"],state["Russian"],state
    with open(audio_path,"rb") as f:
        stt = client.audio.transcriptions.create(
            model="whisper-1", file=f, language=LANG_CODE.get(src))
    full = stt.text.strip()
    new = full[len(state["orig"]):]
    if new:
        state["orig"] = full
        for lang in FOUR:
            state[lang] += " " + _gpt(new, src, lang)
    return (state["orig"].strip(),
            state["English"].strip(),
            state["Chinese"].strip(),
            state["Thai"].strip(),
            state["Russian"].strip(),
            state)

# ────────────── 6. Gradio UI ───────────────────────────────────
with gr.Blocks(title="SMARTok Demo", theme=gr.themes.Soft()) as app:
    with gr.Tabs():
        # 탭 1
        with gr.TabItem("🎙️ 오디오 번역"):
            src1 = gr.Dropdown(LANGUAGES,value="Korean",label="입력 언어")
            tgt1 = gr.Dropdown(LANGUAGES,value="English",label="출력 언어")
            aud1 = gr.Audio(sources=["microphone","upload"],type="filepath")
            res1 = gr.Button("번역")
            o1   = gr.Textbox(label="원문",lines=5)
            t1   = gr.Textbox(label="번역",lines=5)
            a1   = gr.Audio(label="TTS",type="filepath",autoplay=True)
            res1.click(translate_audio,[aud1,src1,tgt1],[o1,t1,a1])

        # 탭 2
        with gr.TabItem("📄 문서·이미지 번역"):
            src2 = gr.Dropdown(LANGUAGES,value="Korean",label="입력 언어")
            tgt2 = gr.Dropdown(LANGUAGES,value="English",label="출력 언어")
            file2= gr.File(label="PDF/이미지 업로드",
                           file_types=[".pdf",".png",".jpg",".jpeg",
                                       ".bmp",".tiff",".gif"])
            doc2 = gr.Button("번역")
            o2   = gr.Textbox(label="추출 원문",lines=15)
            t2   = gr.Textbox(label="번역 결과",lines=15)
            doc2.click(translate_doc,[file2,src2,tgt2],[o2,t2])

        # 탭 3
        with gr.TabItem("⏱️ 실시간 1언어"):
            src3 = gr.Dropdown(LANGUAGES,value="Korean",label="입력 언어")
            tgt3 = gr.Dropdown(LANGUAGES,value="English",label="출력 언어")
            mic3 = gr.Audio(sources=["microphone"],streaming=True)
            o3   = gr.Textbox(label="원문(실시간)",lines=8)
            t3   = gr.Textbox(label="번역(실시간)",lines=8)
            st3  = gr.State()
            mic3.stream(stream_single,
                        inputs=[src3,tgt3,st3],
                        outputs=[o3,t3,st3])

        # 탭 4
        with gr.TabItem("🌏 실시간 4개 언어"):
            src4 = gr.Dropdown(LANGUAGES,value="Korean",label="입력 언어")
            mic4 = gr.Audio(sources=["microphone"],streaming=True)
            o4   = gr.Textbox(label="원문",lines=8)
            e4   = gr.Textbox(label="English",lines=8)
            c4   = gr.Textbox(label="Chinese(简体)",lines=8)
            th4  = gr.Textbox(label="Thai",lines=8)
            r4   = gr.Textbox(label="Russian",lines=8)
            st4  = gr.State()
            mic4.stream(stream_multi,
                        inputs=[src4,st4],
                        outputs=[o4,e4,c4,th4,r4,st4])

# ────────────── 7. 실행 ───────────────────────────────────────
if __name__ == "__main__":
    app.launch(server_name="0.0.0.0",server_port=7860,share=False,debug=True)