app.py

import gradio as gr
import openai
import os
#import torch



#device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
#print("Device type:", device)

#from transformers import pipeline
#from transformers import (
#    AutoTokenizer,
#    WhisperProcessor,
#    WhisperForConditionalGeneration,
#)

api_key = os.getenv('OPEN_API_KEY')
openai.api_key = api_key


global_history = [{"role": "assistant", "content": "Hi, I am a chatbot. I can converse in English. I can answer your questions about farming in India. Ask me anything!"}]

from pydub import AudioSegment

def get_asr_model_processor(model_id):
    processor = WhisperProcessor.from_pretrained(model_id,model_max_length=225)
    model = WhisperForConditionalGeneration.from_pretrained(model_id).to(device)
    # model.forced_decoder_ids =None
    model.config.max_new_token = 200
    return {
        "model": model,
        "processor": processor
    }

 
#model_proc_dict  = get_asr_model_processor("vasista22/whisper-hindi-large-v2")
#asr_pipe = pipeline("automatic-speech-recognition", model=model_proc_dict["model"], tokenizer=model_proc_dict["processor"].tokenizer, feature_extractor=model_proc_dict["processor"].feature_extractor,device=device)




def get_asr_output(audio_path,lang):
    audio = AudioSegment.from_file(audio_path)
    audio.export("temp.wav", format="wav")
    file = open("temp.wav","rb")
    
    
    transcription = openai.Audio.transcribe("whisper-1", file, language=lang)
    op_text = transcription.text

    """ if lang == "hi":
        op_text = asr_pipe("temp.wav")['text']
        print('whisper',transcription)
        print('ai4b',op_text) """
    
    return op_text

def add_text(history, audio_path,lang):
    global global_history
    text = get_asr_output(audio_path,lang_dict[lang])
    history = history + [(text, None)]
    global_history = global_history+[{"role": "user", "content": text}]
    print(global_history)
    return history, ""

def get_chatgpt_response(history):
    output = openai.ChatCompletion.create(model="gpt-3.5-turbo", messages=history)
    history.append({"role": "assistant", "content": output.choices[0].message.content})
    return output.choices[0].message.content, history

def bot(history):
    global global_history
    response, global_history = get_chatgpt_response(global_history)
    history[-1][1] = response
    return history

def clear_history(lang = "English"):
    global global_history
    global_history = [{"role": "assistant", "content": "Hi, I am a chatbot. I can converse in {}. I can answer your questions about farming in India. Ask me anything!".format(lang)}]
    return None

lang_dict = {
    "English": "en",
    "Hindi": "hi",
    "Bengali": "bn",
    "Gujarati": "gu",
    "Kannada": "kn",
    "Marathi": "mr",
    "Tamil": "ta",
    "Telugu": "te"
}

with gr.Blocks(title="Krishi GPT Demo") as demo:
    lang = gr.Radio(list(lang_dict.keys()), label="Select a Language")
    
    
    
    with gr.Row():
        with gr.Column():
            user_audio = gr.Audio(source="microphone",type="filepath",label = "Speak your query")
            txt = gr.Textbox(
                show_label=True,
                placeholder="Enter text and press enter, or Record your Audio",
                visible=False,
            ).style(container=False)
    
    submit = gr.Button("Submit")
    chatbot = gr.Chatbot([], elem_id="chatbot").style(height=500)
    clear = gr.Button("Clear")
    submit.click(add_text, [chatbot, user_audio,lang], [chatbot, txt]).then(bot, chatbot, chatbot)
    clear.click(clear_history, [lang], chatbot, queue=False)
    lang.change(clear_history, [lang], chatbot, queue=False)


demo.launch(share=False)