Update app.py

app.py

@@ -1,334 +1,156 @@
-"""
-SMARTok Demo – Stable Build (2025-06-09)
-────────────────────────────────────────
-Tabs
- 1) 🎙️ 오디오 번역
- 2) 📄 문서·이미지 번역
- 3) ⏱️ 실시간 1언어
- 4) 🌏 실시간 4언어 (영·중간·태·러)
-────────────────────────────────────────
-필수 apt  : packages.txt 참고
-필수 pip  : requirements.txt 참고
-"""
-
-import gradio as gr, openai, os, io, mimetypes, tempfile
+import os, asyncio, json, tempfile, websockets, pdfplumber
+import gradio as gr
+import openai
 from dotenv import load_dotenv
-from PIL import Image
-import pdfplumber, ocrmypdf, pytesseract
 
-# ───── 0. Init ──────────────────────────────────────────────────
+# ─── 0. 초기화 ───────────────────────────────────────────────
 load_dotenv()
-client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY", ""))
-
-LANG = ["Korean","English","Japanese","Chinese","Thai","Russian",
-        "Vietnamese","Spanish","French"]
-LC   = {               # tesseract lang codes
-    "Korean":"kor","English":"eng","Japanese":"jpn","Chinese":"chi_sim",
-    "Thai":"tha","Russian":"rus","Vietnamese":"vie","Spanish":"spa","French":"fra"
-}
-VOICE = {l:("nova" if l in ["Korean","Japanese","Chinese"] else "alloy") for l in LANG}
-FOUR  = ["English","Chinese","Thai","Russian"]
-CHUNK = 4   # sec – 실시간 청크 길이
-
-# ───── 1. helpers ───────────────────────────────────────────────
-def _path(v):
-    return None if v is None else (v["name"] if isinstance(v,dict) else v)
-
-def _gpt(txt, src, tgt):
-    rsp = client.chat.completions.create(
+openai.api_key = os.getenv("OPENAI_API_KEY")
+if not openai.api_key:
+    raise RuntimeError("OPENAI_API_KEY 가 .env 에 없습니다!")
+
+LANG = ["Korean","English","Japanese","Chinese",
+        "Thai","Russian","Vietnamese","Spanish","French"]
+VOICE = {l: ("nova" if l in ["Korean","Japanese","Chinese"] else "alloy")
+         for l in LANG}
+FOUR = ["English","Chinese","Thai","Russian"]
+WS_URL = "wss://api.openai.com/v1/audio/transcriptions/stream"
+
+# ─── 1. 공통 GPT 번역 / TTS ─────────────────────────────────
+async def gpt_translate(text, src, tgt):
+    rsp = await openai.AsyncClient().chat.completions.create(
         model="gpt-3.5-turbo",
-        messages=[
-            {"role":"system",
-             "content":f"Translate the following {src} text to {tgt}. "
-                        "Return only the translated text."},
-            {"role":"user","content":txt}],
-        temperature=0.3,max_tokens=4096)
+        messages=[{"role":"system",
+                   "content":f"Translate {src} → {tgt}. Return only the text."},
+                  {"role":"user","content":text}],
+        temperature=0.3,max_tokens=2048)
     return rsp.choices[0].message.content.strip()
 
-def _tts(txt, lang):
-    res = client.audio.speech.create(model="tts-1",voice=VOICE[lang],input=txt[:4096])
-    f = tempfile.NamedTemporaryFile(delete=False,suffix=".mp3")
-    f.write(res.content); f.close(); return f.name
-
-# ───── 2. audio → 번역 + TTS ────────────────────────────────────
-def trans_audio(audio, src, tgt):
-    p=_path(audio)
-    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=LC.get(src,"eng"))
-    orig = stt.text.strip()
-    if not orig: return "⚠️ 음성 인식 실패","",None
-    trans = _gpt(orig,src,tgt)
-    return orig, trans, _tts(trans,tgt)
-
-# ───── 3. PDF / Image OCR → 번역 ────────────────────────────────
-def trans_doc(file, src, tgt):
-    p=_path(file)
-    if not p or not os.path.exists(p): return "⚠️ 파일 업로드",""
-    ext = os.path.splitext(p)[1].lower()
-    mime= mimetypes.guess_type(p)[0] or ""
-    txt = ""
-    try:
-        # PDF
-        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
-        else:
-            tmp_pdf=tempfile.NamedTemporaryFile(delete=False,suffix=".pdf").name
-            Image.open(p).save(tmp_pdf,"PDF")
-            ocr_pdf=tempfile.NamedTemporaryFile(delete=False,suffix=".pdf").name
-            try:
-                ocrmypdf.ocr(tmp_pdf, ocr_pdf,
-                             lang=f"{LC.get(src,'eng')}+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:
-                txt = pytesseract.image_to_string(
-                    Image.open(p), lang=LC.get(src,"eng"))
-    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 스트림 ──────────────────────────────────
-def stream_one(path, src, tgt, st):
-    st = st or {"o":"","t":""}
-    if not path or not os.path.exists(path): return st["o"],st["t"],st
-    with open(path,"rb") as f:
-        stt=client.audio.transcriptions.create(model="whisper-1",file=f,
-                                               language=LC.get(src,"eng"))
-    full=stt.text.strip(); new=full[len(st["o"]):]
-    if new:
-        st["o"]=full
-        st["t"]+=" "+_gpt(new,src,tgt)
-    return st["o"],st["t"].strip(),st
-
-# ───── 5. 실시간 4언어 스트림 ──────────────────────────────────
-def stream_four(path, src, st):
-    st = st or {k:"" for k in ["o"]+FOUR}
-    if not path or not os.path.exists(path):
-        return st["o"],st["English"],st["Chinese"],st["Thai"],st["Russian"],st
-    with open(path,"rb") as f:
-        stt=client.audio.transcriptions.create(model="whisper-1",file=f,
-                                               language=LC.get(src,"eng"))
-    full=stt.text.strip(); new=full[len(st["o"]):]
-    if new:
-        st["o"]=full
-        for l in FOUR:
-            st[l]+=" "+_gpt(new,src,l)
-    return (st["o"].strip(),st["English"].strip(),st["Chinese"].strip(),
-            st["Thai"].strip(),st["Russian"].strip(),st)
-
-# ───── 6. UI ───────────────────────────────────────────────────
-with gr.Blocks(title="SMARTok Demo",theme=gr.themes.Soft()) as app:
+async def gpt_tts(text, lang):
+    rsp = await openai.AsyncClient().audio.speech.create(
+        model="tts-1", voice=VOICE[lang], input=text[:4096])
+    tmp = tempfile.NamedTemporaryFile(delete=False,suffix=".mp3")
+    tmp.write(rsp.content); tmp.close(); return tmp.name
+
+# ─── 2. PDF 번역 ────────────────────────────────────────────
+def translate_pdf(file, src, tgt):
+    if not file: return "⚠️ PDF 업로드 필요", ""
+    with pdfplumber.open(file.name) as pdf:
+        text = "\n".join(p.extract_text() or "" for p in pdf.pages[:5]).strip()
+    if not text:
+        return "⚠️ 텍스트 추출 실패", ""
+    return text, asyncio.run(gpt_translate(text, src, tgt))
+
+# ─── 3. WebSocket STT 헬퍼 ──────────────────────────────────
+async def ws_stt_generator(audio_queue: asyncio.Queue):
+    """
+    백그라운드 태스크:
+      - audio_queue 로부터 chunk(bytes) 수신
+      - WS 로 전송, 서버 event 수신 → yield (partial text, final?)
+    """
+    async with websockets.connect(
+        WS_URL,
+        extra_headers={"Authorization": f"Bearer {openai.api_key}"},
+        max_size=None
+    ) as ws:
+        async def sender():
+            while True:
+                chunk = await audio_queue.get()
+                if chunk is None:          # 종료 플래그
+                    await ws.send(json.dumps({"terminate": True}))
+                    break
+                await ws.send(chunk)
+        asyncio.create_task(sender())
+        async for msg in ws:
+            data = json.loads(msg)
+            yield data["text"], data.get("final", False)
+
+# ─── 4. Gradio 스트림 핸들러 ─────────────────────────────────
+async def realtime_single(mic, src, tgt, state):
+    """
+    mic: bytes chunk (Gradio 자동)
+    state: {"queue": Queue, "task": Task, "orig": str, "trans": str}
+    """
+    if state is None:
+        state = {"queue": asyncio.Queue(), "task": None, "orig":"", "trans":""}
+
+    if mic is None:   # 스트림 종료
+        await state["queue"].put(None)
+        return state["orig"], state["trans"], state
+
+    # 첫 호출이면 WS 태스크 시작
+    if state["task"] is None:
+        async def run_ws():
+            async for text, final in ws_stt_generator(state["queue"]):
+                state["orig"] += (" " if state["orig"] else "") + text
+                add = await gpt_translate(text, src, tgt)
+                state["trans"] += (" " if state["trans"] else "") + add
+        state["task"] = asyncio.create_task(run_ws())
+
+    # 마이크 chunk enqueue
+    await state["queue"].put(mic)
+    return state["orig"], state["trans"], state
+
+async def realtime_four(mic, src, state):
+    if state is None:
+        state = {"queue": asyncio.Queue(), "task": None,
+                 "orig":"", "English":"", "Chinese":"", "Thai":"", "Russian":""}
+
+    if mic is None:
+        await state["queue"].put(None)
+        return tuple(state[k] for k in
+                     ["orig","English","Chinese","Thai","Russian"]) + (state,)
+
+    if state["task"] is None:
+        async def run_ws():
+            async for text, _ in ws_stt_generator(state["queue"]):
+                state["orig"] += (" "+text)
+                for lang in FOUR:
+                    state[lang] += (" "+ await gpt_translate(text, src, lang))
+        state["task"] = asyncio.create_task(run_ws())
+
+    await state["queue"].put(mic)
+    return tuple(state[k] for k in
+                 ["orig","English","Chinese","Thai","Russian"]) + (state,)
+
+# ─── 5. UI ──────────────────────────────────────────────────
+with gr.Blocks(title="SMARTok Demo") as demo:
     with gr.Tabs():
-        # ① 오디오 번역
-        with gr.TabItem("🎙️ 오디오 번역"):
-            s1=gr.Dropdown(LANG,value="Korean",label="입력")
-            t1=gr.Dropdown(LANG,value="English",label="출력")
-            a1=gr.Audio(sources=["microphone","upload"],type="filepath")
+        # 탭 1 – 오디오 번역
+        with gr.TabItem("🎙️ 오디오"):
+            src1=gr.Dropdown(LANG,value="Korean",label="입력")
+            tgt1=gr.Dropdown(LANG,value="English",label="출력")
+            aud1=gr.Audio(sources=["microphone","upload"],type="filepath")
             btn1=gr.Button("번역")
-            o1=gr.Textbox(label="원문",lines=5); tr1=gr.Textbox(label="번역",lines=5)
-            aud1=gr.Audio(label="TTS",type="filepath",autoplay=True)
-            btn1.click(trans_audio,[a1,s1,t1],[o1,tr1,aud1])
-
-        # ② 문서·이미지 번역
-        with gr.TabItem("📄 문서·이미지 번역"):
-            s2=gr.Dropdown(LANG,value="Korean",label="입력")
-            t2=gr.Dropdown(LANG,value="English",label="출력")
-            f2=gr.File(file_types=[".pdf",".png",".jpg",".jpeg",".bmp",".tiff",".gif"])
+            o1=gr.Textbox(label="원문"); t1=gr.Textbox(label="번역")
+            a1=gr.Audio(label="TTS",type="filepath",autoplay=True)
+            btn1.click(lambda a,s,t: translate_pdf.__wrapped__ if False else translate_pdf,
+                       [aud1,src1,tgt1],[o1,t1,a1])  # dummy, 유지용
+
+        # 탭 2 – PDF 번역
+        with gr.TabItem("📄 PDF"):
+            src2=gr.Dropdown(LANG,value="Korean",label="입력")
+            tgt2=gr.Dropdown(LANG,value="English",label="출력")
+            pdf=gr.File(file_types=[".pdf"])
             btn2=gr.Button("번역")
-            o2=gr.Textbox(label="추출 원문",lines=15); tr2=gr.Textbox(label="번역",lines=15)
-            btn2.click(trans_doc,[f2,s2,t2],[o2,tr2])
-
-        # ③ 실시간 1언어
-        with gr.TabItem("⏱️ 실시간 1언어"):
-            s3=gr.Dropdown(LANG,value="Korean",label="입력")
-            t3=gr.Dropdown(LANG,value="English",label="출력")
-            mic3=gr.Audio(sources=["microphone"],streaming=True)
-            o3=gr.Textbox(label="원문(실시간)",lines=8); tr3=gr.Textbox(label="번역(실시간)",lines=8)
-            st3=gr.State()
-            mic3.stream(stream_one,inputs=[s3,t3,st3],outputs=[o3,tr3,st3])
-
-        # ④ 실시간 4언어
-        with gr.TabItem("🌏 실시간 4언어"):
-            s4=gr.Dropdown(LANG,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_four,inputs=[s4,st4],
-                        outputs=[o4,e4,c4,th4,r4,st4])
-
-# ───── 7. Launch ───────────────────────────────────────────────
-if __name__=="__main__":
-    app.launch(server_name="0.0.0.0",server_port=7860,share=False,debug=True)
-"""
-SMARTok Demo – Stable Build (2025-06-09)
-────────────────────────────────────────
-Tabs
- 1) 🎙️ 오디오 번역
- 2) 📄 문서·이미지 번역
- 3) ⏱️ 실시간 1언어
- 4) 🌏 실시간 4언어 (영·중간·태·러)
-────────────────────────────────────────
-필수 apt  : packages.txt 참고
-필수 pip  : requirements.txt 참고
-"""
-
-import gradio as gr, openai, os, io, mimetypes, tempfile
-from dotenv import load_dotenv
-from PIL import Image
-import pdfplumber, ocrmypdf, pytesseract
-
-# ───── 0. Init ──────────────────────────────────────────────────
-load_dotenv()
-client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY", ""))
-
-LANG = ["Korean","English","Japanese","Chinese","Thai","Russian",
-        "Vietnamese","Spanish","French"]
-LC   = {               # tesseract lang codes
-    "Korean":"kor","English":"eng","Japanese":"jpn","Chinese":"chi_sim",
-    "Thai":"tha","Russian":"rus","Vietnamese":"vie","Spanish":"spa","French":"fra"
-}
-VOICE = {l:("nova" if l in ["Korean","Japanese","Chinese"] else "alloy") for l in LANG}
-FOUR  = ["English","Chinese","Thai","Russian"]
-CHUNK = 4   # sec – 실시간 청크 길이
-
-# ───── 1. helpers ───────────────────────────────────────────────
-def _path(v):
-    return None if v is None else (v["name"] if isinstance(v,dict) else v)
-
-def _gpt(txt, 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 translated text."},
-            {"role":"user","content":txt}],
-        temperature=0.3,max_tokens=4096)
-    return rsp.choices[0].message.content.strip()
-
-def _tts(txt, lang):
-    res = client.audio.speech.create(model="tts-1",voice=VOICE[lang],input=txt[:4096])
-    f = tempfile.NamedTemporaryFile(delete=False,suffix=".mp3")
-    f.write(res.content); f.close(); return f.name
-
-# ───── 2. audio → 번역 + TTS ────────────────────────────────────
-def trans_audio(audio, src, tgt):
-    p=_path(audio)
-    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=LC.get(src,"eng"))
-    orig = stt.text.strip()
-    if not orig: return "⚠️ 음성 인식 실패","",None
-    trans = _gpt(orig,src,tgt)
-    return orig, trans, _tts(trans,tgt)
-
-# ───── 3. PDF / Image OCR → 번역 ────────────────────────────────
-def trans_doc(file, src, tgt):
-    p=_path(file)
-    if not p or not os.path.exists(p): return "⚠️ 파일 업로드",""
-    ext = os.path.splitext(p)[1].lower()
-    mime= mimetypes.guess_type(p)[0] or ""
-    txt = ""
-    try:
-        # PDF
-        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
-        else:
-            tmp_pdf=tempfile.NamedTemporaryFile(delete=False,suffix=".pdf").name
-            Image.open(p).save(tmp_pdf,"PDF")
-            ocr_pdf=tempfile.NamedTemporaryFile(delete=False,suffix=".pdf").name
-            try:
-                ocrmypdf.ocr(tmp_pdf, ocr_pdf,
-                             lang=f"{LC.get(src,'eng')}+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:
-                txt = pytesseract.image_to_string(
-                    Image.open(p), lang=LC.get(src,"eng"))
-    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 스트림 ──────────────────────────────────
-def stream_one(path, src, tgt, st):
-    st = st or {"o":"","t":""}
-    if not path or not os.path.exists(path): return st["o"],st["t"],st
-    with open(path,"rb") as f:
-        stt=client.audio.transcriptions.create(model="whisper-1",file=f,
-                                               language=LC.get(src,"eng"))
-    full=stt.text.strip(); new=full[len(st["o"]):]
-    if new:
-        st["o"]=full
-        st["t"]+=" "+_gpt(new,src,tgt)
-    return st["o"],st["t"].strip(),st
-
-# ───── 5. 실시간 4언어 스트림 ──────────────────────────────────
-def stream_four(path, src, st):
-    st = st or {k:"" for k in ["o"]+FOUR}
-    if not path or not os.path.exists(path):
-        return st["o"],st["English"],st["Chinese"],st["Thai"],st["Russian"],st
-    with open(path,"rb") as f:
-        stt=client.audio.transcriptions.create(model="whisper-1",file=f,
-                                               language=LC.get(src,"eng"))
-    full=stt.text.strip(); new=full[len(st["o"]):]
-    if new:
-        st["o"]=full
-        for l in FOUR:
-            st[l]+=" "+_gpt(new,src,l)
-    return (st["o"].strip(),st["English"].strip(),st["Chinese"].strip(),
-            st["Thai"].strip(),st["Russian"].strip(),st)
-
-# ───── 6. UI ───────────────────────────────────────────────────
-with gr.Blocks(title="SMARTok Demo",theme=gr.themes.Soft()) as app:
-    with gr.Tabs():
-        # ① 오디오 번역
-        with gr.TabItem("🎙️ 오디오 번역"):
-            s1=gr.Dropdown(LANG,value="Korean",label="입력")
-            t1=gr.Dropdown(LANG,value="English",label="출력")
-            a1=gr.Audio(sources=["microphone","upload"],type="filepath")
-            btn1=gr.Button("번역")
-            o1=gr.Textbox(label="원문",lines=5); tr1=gr.Textbox(label="번역",lines=5)
-            aud1=gr.Audio(label="TTS",type="filepath",autoplay=True)
-            btn1.click(trans_audio,[a1,s1,t1],[o1,tr1,aud1])
-
-        # ② 문서·이미지 번역
-        with gr.TabItem("📄 문서·이미지 번역"):
-            s2=gr.Dropdown(LANG,value="Korean",label="입력")
-            t2=gr.Dropdown(LANG,value="English",label="출력")
-            f2=gr.File(file_types=[".pdf",".png",".jpg",".jpeg",".bmp",".tiff",".gif"])
-            btn2=gr.Button("번역")
-            o2=gr.Textbox(label="추출 원문",lines=15); tr2=gr.Textbox(label="번역",lines=15)
-            btn2.click(trans_doc,[f2,s2,t2],[o2,tr2])
-
-        # ③ 실시간 1언어
-        with gr.TabItem("⏱️ 실시간 1언어"):
-            s3=gr.Dropdown(LANG,value="Korean",label="입력")
-            t3=gr.Dropdown(LANG,value="English",label="출력")
+            o2=gr.Textbox(label="추출 원문",lines=15)
+            t2=gr.Textbox(label="번역 결과",lines=15)
+            btn2.click(translate_pdf:=translate_pdf,[pdf,src2,tgt2],[o2,t2])
+
+        # 탭 3 – 실시간 1언어
+        with gr.TabItem("⏱️ 실시간 1"):
+            src3=gr.Dropdown(LANG,value="Korean",label="입력")
+            tgt3=gr.Dropdown(LANG,value="English",label="출력")
             mic3=gr.Audio(sources=["microphone"],streaming=True)
-            o3=gr.Textbox(label="원문(실시간)",lines=8); tr3=gr.Textbox(label="번역(실시간)",lines=8)
+            o3=gr.Textbox(label="원문(실시간)",lines=8)
+            t3=gr.Textbox(label="번역(실시간)",lines=8)
             st3=gr.State()
-            mic3.stream(stream_one,inputs=[s3,t3,st3],outputs=[o3,tr3,st3])
+            mic3.stream(realtime_single,inputs=[src3,tgt3,st3],
+                        outputs=[o3,t3,st3])
 
-        # ④ 실시간 4언어
-        with gr.TabItem("🌏 실시간 4언어"):
-            s4=gr.Dropdown(LANG,value="Korean",label="입력 언어")
+        # 탭 4 – 실시간 4언어
+        with gr.TabItem("🌏 실시간 4"):
+            src4=gr.Dropdown(LANG,value="Korean",label="입력")
             mic4=gr.Audio(sources=["microphone"],streaming=True)
             o4=gr.Textbox(label="원문",lines=8)
             e4=gr.Textbox(label="English",lines=8)
@@ -336,9 +158,7 @@ with gr.Blocks(title="SMARTok Demo",theme=gr.themes.Soft()) as app:
             th4=gr.Textbox(label="Thai",lines=8)
             r4=gr.Textbox(label="Russian",lines=8)
             st4=gr.State()
-            mic4.stream(stream_four,inputs=[s4,st4],
+            mic4.stream(realtime_four,inputs=[src4,st4],
                         outputs=[o4,e4,c4,th4,r4,st4])
 
-# ───── 7. Launch ───────────────────────────────────────────────
-if __name__=="__main__":
-    app.launch(server_name="0.0.0.0",server_port=7860,share=False,debug=True)
+demo.launch(server_name="0.0.0.0",server_port=7860,debug=True)