commit

app.py

@@ -1,18 +1,7 @@
-# app.py – Prüfungsrechts-Chatbot (RAG + Sprachmodus)
-# Version 26.11 – ohne Modi, stabil für Text + Voice
+# app.py – Prüfungsrechts-Chatbot (RAG + Sprache, UI kiểu ChatGPT)
 
 import gradio as gr
 from gradio_pdf import PDF
-from huggingface_hub import hf_hub_download
-
-# from load_documents import load_documents, DATASET, PDF_FILE, HTML_FILE
-# from split_documents import split_documents
-# from vectorstore import build_vectorstore
-# from retriever import get_retriever
-# from llm import load_llm
-# from rag_pipeline import answer, PDF_BASE_URL, LAW_URL
-
-# from speech_io import transcribe_audio, synthesize_speech
 
 from load_documents import load_all_documents
 from split_documents import split_documents
@@ -41,10 +30,10 @@ retriever = get_retriever(vs)
 print("🤖 Lade LLM…")
 llm = load_llm()
 
+
 # =====================================================
 # Quellen formatieren – Markdown für Chat
 # =====================================================
-
 def format_sources(src):
     if not src:
         return ""
@@ -53,140 +42,165 @@ def format_sources(src):
 
     for s in src:
         line = f"- [{s['source']}]({s['url']})"
-        if s.get("page"):
+        if s.get("page") is not None:
             line += f" (Seite {s['page']})"
         out.append(line)
 
     return "\n".join(out)
 
-# ===================== TEXT CHAT =====================
-
-def chatbot_text(msg, history):
-    if not msg:
-        return history, ""
-
-    ans, sources = answer(msg, retriever, llm)
-    history.append({"role": "user", "content": msg})
-    history.append({"role": "assistant", "content": ans + format_sources(sources)})
-    return history, ""
-
-# ===================== VOICE CHAT =====================
-
-def chatbot_voice(audio, history):
-    text = transcribe_audio(audio)
-    history.append({"role": "user", "content": text})
 
+# =====================================================
+# CORE CHAT-FUNKTION (MultimodalTextbox: Text + Audio)
+# =====================================================
+def chat_fn(message, history):
+    """
+    message: dict {"text": str, "files": [...]} von gr.MultimodalTextbox
+    history: Liste von OpenAI-ähnlichen Messages (role, content)
+    """
+    # 1) Text + evtl. Audio aus message holen
+    if isinstance(message, dict):
+        text = (message.get("text") or "").strip()
+        files = message.get("files") or []
+    else:
+        text = str(message or "").strip()
+        files = []
+
+    # Audio-Datei (vom Mikrofon) herausziehen
+    audio_path = None
+    for f in files:
+        # gr.MultimodalTextbox liefert i.d.R. Dict mit "path"
+        if isinstance(f, dict):
+            path = f.get("path")
+        else:
+            path = f
+        if isinstance(path, str) and path:
+            audio_path = path
+            break
+
+    # Wenn Audio vorhanden: transkribieren
+    if audio_path:
+        spoken = transcribe_audio(audio_path)
+        if text:
+            text = (text + " " + spoken).strip()
+        else:
+            text = spoken
+
+    if not text:
+        # Nichts zu tun
+        return history, None, {"text": "", "files": []}
+
+    # 2) RAG-Antwort berechnen
     ans, sources = answer(text, retriever, llm)
     bot_msg = ans + format_sources(sources)
 
-    history.append({"role": "assistant", "content": bot_msg})
+    # 3) History aktualisieren (ChatGPT-Style)
+    history = history + [
+        {"role": "user", "content": text},
+        {"role": "assistant", "content": bot_msg},
+    ]
 
+    # 4) TTS für Antwort
     tts_audio = synthesize_speech(bot_msg)
-    return history, tts_audio, ""
+
+    # 5) Input-Feld leeren
+    cleared_input = {"text": "", "files": []}
+
+    return history, tts_audio, cleared_input
 
 
 # =====================================================
-# LAST ANSWER → TTS
+# LAST ANSWER → TTS (für Button "Antwort erneut vorlesen")
 # =====================================================
-
 def read_last_answer(history):
     if not history:
         return None
 
     for msg in reversed(history):
-        if msg["role"] == "assistant":
-            return synthesize_speech(msg["content"])
+        if msg.get("role") == "assistant":
+            return synthesize_speech(msg.get("content", ""))
 
     return None
 
+
 # =====================================================
 # UI – GRADIO
 # =====================================================
-
 with gr.Blocks(title="Prüfungsrechts-Chatbot (RAG + Sprache)") as demo:
     gr.Markdown("# 🧑‍⚖️ Prüfungsrechts-Chatbot")
     gr.Markdown(
         "Dieser Chatbot beantwortet Fragen **ausschließlich** aus der "
-        "Prüfungsordnung (PDF) und dem Hochschulgesetz NRW (Website). "
+        "Prüfungsordnung (PDF) und dem Hochschulgesetz NRW. "
         "Du kannst Text eingeben oder direkt ins Mikrofon sprechen."
     )
 
     with gr.Row():
+        # ===================== LINKER TEIL: Chat =====================
         with gr.Column(scale=2):
-            chatbot = gr.Chatbot(label="Chat", height=500)
-
-            msg = gr.Textbox(
-                label="Frage eingeben",
-                placeholder="Stelle deine Frage zum Prüfungsrecht …",
+            chatbot = gr.Chatbot(
+                label="Chat",
+                height=500,
+                type="messages",  # nutzt role/content-Struktur
             )
 
-            # TEXT SENDEN
-            msg.submit(
-                chatbot_text,
-                [msg, chatbot],
-                [chatbot, msg]
-            )
+            # Audio-Ausgabe (TTS)
+            voice_out = gr.Audio(label="Vorgelesene Antwort", type="numpy")
 
-            send_btn = gr.Button("Senden (Text)")
-            send_btn.click(
-                chatbot_text,
-                [msg, chatbot],
-                [chatbot, msg]
+            # Multimodal-Textbox mit Mikrofon in der Leiste
+            chat_input = gr.MultimodalTextbox(
+                label=None,
+                placeholder="Stelle deine Frage zum Prüfungsrecht … oder sprich ins Mikrofon",
+                show_label=False,
+                sources=["microphone"],  # nur Mikrofon (kein Upload nötig)
+                file_types=["audio"],
+                max_lines=6,
             )
 
-            # SPRACHEINGABE
-            gr.Markdown("### 🎙️ Spracheingabe")
-            voice_in = gr.Audio(sources=["microphone"], type="filepath")
-            voice_out = gr.Audio(label="Vorgelesene Antwort", type="numpy")
+            # Senden bei Enter / Klick auf Icon
+            chat_input.submit(
+                chat_fn,
+                [chat_input, chatbot],
+                [chatbot, voice_out, chat_input],
+            )
 
-            voice_btn = gr.Button("Sprechen & senden")
-            voice_btn.click(
-                chatbot_voice,
-                [voice_in, chatbot],
-                [chatbot, voice_out, msg]
+            send_btn = gr.Button("Senden")
+            send_btn.click(
+                chat_fn,
+                [chat_input, chatbot],
+                [chatbot, voice_out, chat_input],
             )
 
+            # Button: Antwort erneut vorlesen
             read_btn = gr.Button("🔁 Antwort erneut vorlesen")
             read_btn.click(
                 read_last_answer,
                 [chatbot],
-                [voice_out]
+                [voice_out],
             )
 
+            # Chat löschen
             clear_btn = gr.Button("Chat zurücksetzen")
-            clear_btn.click(lambda: [], None, chatbot)
-
-        # =====================
-        # RECHTE SPALTE: Viewer
-        # =====================
-
-        # with gr.Column(scale=1):
-        #     gr.Markdown("### 📄 Prüfungsordnung (PDF)")
-        #     PDF(_pdf_path, height=350)
-
-        #     gr.Markdown("### 📘 Hochschulgesetz NRW (Website)")
-        #     gr.HTML(
-        #         f'<iframe src="{LAW_URL}" style="width:100%;height:350px;border:none;"></iframe>'
-        #     )
+            clear_btn.click(
+                lambda: ([], None, {"text": "", "files": []}),
+                None,
+                [chatbot, voice_out, chat_input],
+            )
 
+        # ===================== RECHTER TEIL: Viewer =====================
         with gr.Column(scale=1):
-            gr.Markdown("### 📄 Prüfungsordnung (PDF)")
-            # PDF đã được load_documents cung cấp pdf_url — dùng metadata trực tiếp
+            # PDF-URL aus metadata holen
             pdf_meta = next(d.metadata for d in docs if d.metadata["type"] == "pdf")
+            gr.Markdown("### 📄 Prüfungsordnung (PDF)")
             PDF(pdf_meta["pdf_url"], height=350)
 
-            gr.Markdown("### 📘 Hochschulgesetz NRW")
+            # HG-Viewer-URL (hg_clean.html aus Supabase Storage)
             hg_meta = next(d.metadata for d in docs if d.metadata["type"] == "hg")
-            # hg_view_url = hg_meta["viewer_url"].split("#")[0]
+            hg_url = hg_meta["viewer_url"].split("#")[0]
 
-            hg_url = hg_meta["viewer_url"]
+            gr.Markdown("### 📘 Hochschulgesetz NRW (Viewer)")
             gr.HTML(
                 f'<iframe src="{hg_url}" '
                 'style="width:100%;height:350px;border:none;"></iframe>'
             )
 
-
 if __name__ == "__main__":
     demo.queue().launch(ssr_mode=False, show_error=True)
-
-