Update app.py

app.py

@@ -2,22 +2,47 @@ import gradio as gr
 from transformers import pipeline
 
 asr_pipe = pipeline("automatic-speech-recognition", model="openai/whisper-tiny")
-llm_pipe = pipeline("text-generation", model="microsoft/Phi-3-mini-4k-instruct", trust_remote_code=True)
-tts_pipe = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs")
+llm_pipe = pipeline(
+    "text-generation",
+    model="microsoft/Phi-3-mini-4k-instruct",
+    device_map="auto",      # needs `pip install accelerate`; remove to run on CPU only
+    torch_dtype="auto",
+)
+tts_pipe = pipeline("text-to-speech", model="facebook/mms-tts-eng")
 
-def process_audio(audio):
-    # Convert audio to text using ASR pipeline
-    transcribed_text = asr_pipe(audio)[0]['transcription']
+def process_audio(audio_path):
+    if audio_path is None:
+        return "Please record something first.", None
 
-    # Generate text using LLM pipeline
-    generated_text = llm_pipe(transcribed_text)[0]['generated_text']
+    # 1. Speech -> text
+    user_text = asr_pipe(audio_path)["text"].strip()
 
-    # Convert generated text to speech using TTS pipeline
-    speech = tts_pipe(generated_text)[0]['audio']
+    # 2. Text -> reply (chat format, reply only, capped length)
+    messages = [
+        {"role": "system", "content": "You are a helpful voice assistant. Answer in 1-3 short sentences."},
+        {"role": "user", "content": user_text},
+    ]
+    reply = llm_pipe(
+        messages,
+        max_new_tokens=100,
+        return_full_text=False,
+        do_sample=False,
+    )[0]["generated_text"].strip()
 
-    return speech
+    # 3. Reply -> speech
+    speech = tts_pipe(reply)
+    audio = speech["audio"].squeeze()
+    return reply, (speech["sampling_rate"], audio)
 
-audio_input = gr.Audio(label="Audio Input", sources="microphone")
-my_alexa_output = gr.Interface(fn=process_audio, inputs=audio_input, outputs="text", title="MyAlexa AI Assistant", 
-                      description="An AI Assistant That Will Answer Your Questions!")
-my_alexa_output.launch(inbrowser=True, debug=True)
+demo = gr.Interface(
+    fn=process_audio,
+    inputs=gr.Audio(label="Audio Input", sources=["microphone"], type="filepath"),
+    outputs=[
+        gr.Textbox(label="Assistant reply"),
+        gr.Audio(label="Spoken reply", type="numpy"),
+    ],
+    title="MyAlexa AI Assistant",
+    description="An AI Assistant That Will Answer Your Questions!",
+)
+
+demo.launch(inbrowser=True, debug=True)