Update app.py

app.py

@@ -1,48 +1,37 @@
 import gradio as gr
-from transformers import pipeline
 import torch
+from transformers import pipeline
 
-# Gerät auswählen (GPU, falls verfügbar, sonst CPU)
+# Laden der Modelle (einmalig beim Start)
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-# Lade das kleinere Modell
-pipe = pipeline("text-generation", model="google/flan-t5-xxl", device=device)
-
-def chat(message, history):
-  prompt = f"User: {message}\nAssistant:"
-  sequences = pipe(prompt, max_length=256)
-  response = sequences[0]['generated_text'].split("Assistant:")[1].strip()
-  history.append((message, response))
-  return history, history
-
-def transcribe_and_send(audio, history):
-  if audio is None:
-    return history, "Keine Audioaufnahme erhalten."
-
-  try:
-      asr = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=device)
-      transcription = asr(audio)["text"]
-      return chat(transcription, history)
-  except Exception as e:
-      print(f"Fehler bei der Transkription: {e}")
-      return history, "Fehler bei der Audioverarbeitung."
-
-with gr.Blocks() as demo:
-    chatbot = gr.Chatbot()
-    state = gr.State([])
-
-    with gr.Row():
-        audio_input = gr.Audio(type="filepath", label="Sprachaufnahme")
-        text_input = gr.Textbox(placeholder="Nachricht eingeben...")
-
-    text_output = gr.Textbox()
-    send_button = gr.Button("Senden")
-    clear_button = gr.Button("Chat löschen")
-
-    send_button.click(chat, [text_input, state], [state, chatbot])
-   #audio_input.submit(transcribe_and_send, [audio_input, state], [state, chatbot])
-    clear_button.click(lambda: [], outputs=text_output)
-    #text_input.submit(chat, [text_input, state], [state, chatbot])
+speech_to_text = pipeline("automatic-speech-recognition", model="openai/whisper-small", device=device)
+text_to_speech = pipeline("text-to-speech", model="facebook/fastspeech2-en-ljspeech", device=device)
+
+def audio_to_audio_chatbot(audio):
+    if audio is None:
+        return None, "Bitte eine Audio-Datei hochladen."
+
+    # 1. Speech-to-Text
+    text = speech_to_text(audio)["text"]
+    print(f"User: {text}")
+
+    # 2. Text-to-Text (Hier wird ein einfacher Echo-Bot verwendet, kann durch ein komplexeres Modell ersetzt werden)
+    response_text = f"Du hast gesagt: {text}"
+    print(f"Bot: {response_text}")
+
+    # 3. Text-to-Speech
+    speech = text_to_speech(response_text)
+    return speech["audio"], response_text
 
 if __name__ == "__main__":
-    demo.launch()
+    iface = gr.Interface(
+        fn=audio_to_audio_chatbot,
+        inputs=gr.Audio(source="microphone", type="filepath"),
+        outputs=[gr.Audio(), gr.Textbox()],
+        title="Audio-zu-Audio-Chatbot (Streaming)",
+        description="Spreche in das Mikrofon und der Bot antwortet mit einer Audio-Ausgabe.",
+        live=True  # Aktiviert Streaming
+    )
+
+    iface.launch()