import time import torch import requests import gradio as gr from transformers import pipeline from usellm import Message, Options, UseLLM from transformers import AutoTokenizer, AutoModelForCausalLM from transformers import WhisperProcessor, WhisperForConditionalGeneration def text_to_speech(text_input): CHUNK_SIZE = 1024 url = "https://api.elevenlabs.io/v1/text-to-speech/TxGEqnHWrfWFTfGW9XjX" headers = { "Accept": "audio/mpeg", "Content-Type": "application/json", "xi-api-key": "7f91dfdd5390bbfd9d44148c59644039" } data = { "text": text_input, "model_id": "eleven_monolingual_v1" } audio_write_path = f"""output_{int(time.time())}.mp3""" response = requests.post(url, json=data, headers=headers) with open(audio_write_path, 'wb') as f: for chunk in response.iter_content(chunk_size=CHUNK_SIZE): if chunk: f.write(chunk) return audio_write_path def whisper_inference(input_audio): processor1 = WhisperProcessor.from_pretrained("openai/whisper-large-v2") model1 = WhisperForConditionalGeneration.from_pretrained("openai/whisper-large-v2") forced_decoder_ids = processor1.get_decoder_prompt_ids(task="translate") input_features = processor1(input_audio, sampling_rate=16000, return_tensors="pt").input_features predicted_ids = model1.generate(input_features, forced_decoder_ids=forced_decoder_ids) transcription = processor1.batch_decode(predicted_ids, skip_special_tokens=True) return transcription def biogpt_large_infer(input_text): tokenizer1 = AutoTokenizer.from_pretrained("microsoft/BioGPT-Large-PubMedQA", add_special_tokens=False) model = AutoModelForCausalLM.from_pretrained("microsoft/BioGPT-Large-PubMedQA")#.to('cuda:0') generator = pipeline("text-generation", model=model, tokenizer=tokenizer1)#, device="cuda:0") output = generator(input_text, min_length=100,max_length=1024,num_beams=5,early_stopping=True, num_return_sequences=1, do_sample=True) output = output[0]['generated_text'] output = output.replace('▃','').replace('FREETEXT','').replace('TITLE','').replace('PARAGRAPH','').replace('ABSTRACT','').replace('<','').replace('>','').replace('/','').strip() return output def chatgpt_infer(input_text): # Initialize the service service = UseLLM(service_url="https://usellm.org/api/llm") # Prepare the conversation messages = [ Message(role="system", content="You are a medical assistant, which answers the query based on factual medical information only."), Message(role="user", content=f"Give me few points on the disease {input_text} and its treatment."), ] options = Options(messages=messages) # Interact with the service response = service.chat(options) return response.content def audio_interface_demo(input_audio): en_prompt = whisper_inference(input_audio) biogpt_output = biogpt_large_infer(en_prompt) chatgpt_output = chatgpt_infer(en_prompt) bio_audio_output = text_to_speech(biogpt_output) chat_audio_output = text_to_speech(chatgpt_output) return biogpt_output, chatgpt_output, bio_audio_output, chat_audio_output def text_interface_demo(input_text): #en_prompt = whisper_inference(input_audio) biogpt_output = biogpt_large_infer(input_text) chatgpt_output = chatgpt_infer(input_text) return biogpt_output, chatgpt_output examples = [ ["Meningitis is"], ["Brain Tumour is"] ] app = gr.Blocks() with app: gr.Markdown("# **

Voice based Medical Informational Bot

**") with gr.Row(): with gr.Column(): with gr.Tab("Text"): input_text = gr.Textbox(lines=3, value="Brain Tumour is", label="Text") text_button = gr.Button(value="Predict") with gr.Tab("Audio"): input_audio = gr.Audio(value="input.mp3", source="upload", type="filepath", label='Audio') audio_button = gr.Button(value="Predict") with gr.Row(): with gr.Column(): with gr.Tab("Output Text"): biogpt_output = gr.Textbox(lines=3, label="BioGpt Output") chatgpt_output = gr.Textbox(lines=3,label="ChatGPT Output") with gr.Tab("Output Audio"): biogpt_output = gr.Textbox(lines=3, label="BioGpt Output") chatgpt_output = gr.Textbox(lines=3,label="ChatGPT Output") audio_output1 = gr.Audio(value=None, label="ChatGPT Audio Output") audio_output2 = gr.Audio(value=None, label="BioGpt Audio Output") #gr.Examples(examples, inputs=[input_text], outputs=[prompt_text, output_text, translated_text], fn=biogpt_text, cache_examples=False) text_button.click(text_interface_demo, inputs=[input_text], outputs=[biogpt_output, chatgpt_output]) audio_button.click(audio_interface_demo, inputs=[input_audio], outputs=[biogpt_output, chatgpt_output, audio_output2, audio_output1]) app.launch(debug=True)