Update app.py

app.py

@@ -1,46 +1,46 @@
 import gradio as gr
-from transformers import AutoModelForSpeechSeq2Seq, AutoTokenizer, AutoModelForCausalLM, AutoModelForTextToWaveform
-
-asr_tokenizer = AutoTokenizer.from_pretrained("openai/whisper-tiny")
-asr_model = AutoModelForSpeechSeq2Seq.from_pretrained("openai/whisper-tiny")
-
-llm_tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct", trust_remote_code=True)
-llm_model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-3-mini-4k-instruct", trust_remote_code=True)
-
-tts_tokenizer = AutoTokenizer.from_pretrained("kakao-enterprise/vits-ljs")
-tts_model = AutoModelForTextToWaveform.from_pretrained("kakao-enterprise/vits-ljs")
-
-def asr_transcript(audio):
-    inputs = asr_tokenizer(audio, return_tensors="pt", padding=True, truncation=True)
-    with torch.no_grad():
-        logits = asr_model(input_values=inputs.input_values).logits
-    predicted_ids = torch.argmax(logits, dim=-1)
-    transcription = asr_tokenizer.batch_decode(predicted_ids)[0]
-    return transcription
-
-def llm_response(input_text):
-    input_text_encoded = llm_tokenizer(input_text, return_tensors="pt", padding=True, truncation=True)
-    with torch.no_grad():
-        output_ids = llm_model.generate(input_text_encoded.input_ids, max_length=100, num_return_sequences=1)
-    response = llm_tokenizer.decode(output_ids[0], skip_special_tokens=True)
-    return response
-
-def tts_convert(text):
-    input_ids = tts_tokenizer(text, return_tensors="pt").input_ids
-    with torch.no_grad():
-        audio = tts_model.generate(input_ids)
-    return audio
-
-# Create Gradio interfaces
-audio_input = gr.Audio(source="microphone", label="Speak Here")
-text_output = gr.Textbox(label="Transcription")
-
-text_input = gr.Textbox(label="Enter Text")
-text_output_gen = gr.Textbox(label="Response")
-
-output_audio = gr.Audio(label="Speech Output")
-
-# Combine everything into Gradio interfaces
-gr.Interface(asr_transcript, audio_input, text_output, title="ASR Transcription").launch()
-gr.Interface(llm_response, text_input, text_output_gen, title="Text Generation").launch()
-gr.Interface(tts_convert, text_input, output_audio, title="TTS Speech Output").launch()
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import pipeline
+
+# Load models
+whisper = AutoModelForCausalLM.from_pretrained("openai/whisper-tiny")
+whisper_tokenizer = AutoTokenizer.from_pretrained("openai/whisper-tiny")
+
+llm_model = AutoModelForCausalLM.from_pretrained(
+        "microsoft/Phi-3-mini-4k-instruct", 
+        device_map="cpu", 
+        torch_dtype="auto", 
+        trust_remote_code=True)
+llm_tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3-mini-4k-instruct")
+phi3_mini_4k = pipeline("text-generation", model=llm_model, tokenizer=llm_tokenizer)
+
+vits_tts = pipeline("text-to-speech", model="kakao-enterprise/vits-ljs")
+
+# Define function to convert audio to text
+def audio_to_text(audio):
+    # Convert audio to text using Whisper model
+    inputs = whisper_tokenizer(audio, return_tensors="pt", padding=True, truncation=True)
+    outputs = whisper(**inputs)
+    predicted_ids = torch.argmax(outputs.logits, dim=-1)
+    transcribed_text = whisper_tokenizer.decode(predicted_ids[0])
+    return str(transcribed_text)
+
+# Define functions for model predictions
+def generate_text(prompt):
+    generated_text = phi3_mini_4k(prompt)[0]['generated_text']
+    return generated_text
+
+def text_to_speech(text):
+    generated_text = phi3_mini_4k(text, max_length=50, do_sample=True)[0]['generated_text']
+    speech = vits_tts(generated_text)[0]['audio']
+    return speech
+
+# Create Gradio interface
+audio_input = gr.Audio(label="Can I Help You?", sources="microphone")
+text_output = gr.Textbox(label="Transcribed Text")
+gr.Interface(
+    text_to_speech, 
+    inputs=audio_input, 
+    outputs=gr.Audio(label="Answer:"),
+    title="MyAlexa AI Assistant").launch(inbrowser=True, debug=True)