Update app.py

app.py

@@ -7,6 +7,7 @@ llm_pipe = pipeline(
     model="microsoft/Phi-3-mini-4k-instruct",
     device_map="auto",      # needs `pip install accelerate`; remove to run on CPU only
     torch_dtype="auto",
+    trust_remote_code=True, # Phi-3 needs this on some transformers versions
 )
 tts_pipe = pipeline("text-to-speech", model="facebook/mms-tts-eng")
 
@@ -22,12 +23,14 @@ def process_audio(audio_path):
         {"role": "system", "content": "You are a helpful voice assistant. Answer in 1-3 short sentences."},
         {"role": "user", "content": user_text},
     ]
-    reply = llm_pipe(
+    output = llm_pipe(
         messages,
         max_new_tokens=100,
-        return_full_text=False,
         do_sample=False,
-    )[0]["generated_text"].strip()
+    )
+    # Chat-formatted input returns the full conversation in "generated_text";
+    # the assistant's reply is the last message in that list.
+    reply = output[0]["generated_text"][-1]["content"].strip()
 
     # 3. Reply -> speech
     speech = tts_pipe(reply)