app.py

'''
This script calls the ada model from openai api to predict the next few words.
'''
import os
os.system("pip install --upgrade pip")
from pprint import pprint
os.system("pip install git+https://github.com/openai/whisper.git")
import sys
print("Sys: ", sys.executable)
os.system("pip install openai")
import openai
import gradio as gr
import whisper
from transformers import pipeline
import torch
from transformers import AutoModelForCausalLM
from transformers import AutoTokenizer
import time
# import streaming.py
# from next_word_prediction import GPT2




#gpt2 = AutoModelForCausalLM.from_pretrained("gpt2", return_dict_in_generate=True)
#tokenizer = AutoTokenizer.from_pretrained("gpt2")

### /code snippet


# get gpt2 model
#generator = pipeline('text-generation', model='gpt2')

# whisper model specification 
model = whisper.load_model("tiny")


        
def inference(audio, state=""):

    #time.sleep(2)
    #text = p(audio)["text"]
    #state += text + " "
    # load audio data
    audio = whisper.load_audio(audio)
    # ensure sample is in correct format for inference
    audio = whisper.pad_or_trim(audio)

    # generate a log-mel spetrogram of the audio data
    mel = whisper.log_mel_spectrogram(audio).to(model.device)
    
    _, probs = model.detect_language(mel)

    # decode audio data
    options = whisper.DecodingOptions(fp16 = False)
    # transcribe speech to text
    result = whisper.decode(model, mel, options)
    print("result pre gp model from whisper: ", result, ".text ", result.text, "and the data type: ", type(result.text))

    PROMPT = """The following is an incomplete transcript of a brief conversation. Predict a list of the next most probable words to complete the sentence. 
Some examples: 
Transcript1: Tomorrow night we're going out to 
Predictions1: the movies, a restaurant, a baseball game, the theater, a party for a friend
Transcript2: I would like to order a cheeseburger with a side of
Predictions2: french fries, milkshake, apple slices, salad, extra catsup 
Transcript3: My friend Savanah is
Predictions3: an electrical engineer, a marine biologist, a classical musician 
Transcript4: I need to buy a birthday
Predictions4: present, gift, cake, card

Transcript5: """
    text = PROMPT + result.text + "Prediction5: "
    
    openai.api_key = os.environ["Openai_APIkey"]
    
    response = openai.Completion.create(
                        model="text-ada-001",
                        #model="text-curie-001",
                        prompt=text,
                        temperature=1,
                        max_tokens=8,
                        n=5)

    infers = []
    temp = []
    infered=[]
    for i in range(5):
        print("print1 ", response['choices'][i]['text'])
        temp.append(response['choices'][i]['text'])
        print("print2: infers ", infers)
        print("print3: Responses ", response)
        print("Object type of response: ", type(response))
        #infered = list(map(lambda x: x.split(',')[0], infers))
        #print("Infered type is: ", type(infered))
        infers = list(map(lambda x: x.replace("\n", ""), temp))
        #infered = list(map(lambda x: x.split(','), infers))

        
        

    # result.text
    #return getText, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
    return result.text, state, infers



# get audio from microphone 

gr.Interface(
        fn=inference, 
    inputs=[
        gr.inputs.Audio(source="microphone", type="filepath"), 
        "state"
    ],
    outputs=[
        "textbox",
        "state",
        "textbox"
    ],
    live=True).launch()