app.py

import os
import gradio as gr
# import openai
from openai import OpenAI
from dotenv import load_dotenv
from pathlib import Path
from time import sleep
import audioread
from playsound3 import playsound

load_dotenv(override=True)
key = os.getenv('OPENAI_API_KEY')
users = os.getenv('LOGNAME')
unames = users.split(',')
pwds = os.getenv('PASSWORD')
pwdList = pwds.split(',')

site = os.getenv('SITE')
if site == 'local':
    dp = Path('./data')
    dp.mkdir(exist_ok=True)
    dataDir = './data/'
else:
    dp = Path('/data')
    dp.mkdir(exist_ok=True)
    dataDir = '/data/'

client = OpenAI(api_key = key)

def genUsageStats(do_reset=False):
    result = []
    ttotal4o_in = 0
    ttotal4o_out = 0
    ttotal4mini_in = 0
    ttotal4mini_out = 0
    totalAudio = 0
    for user in unames:
        tokens4o_in = 0
        tokens4o_out = 0
        tokens4mini_in = 0
        tokens4mini_out = 0
        fp = dataDir + user + '_log.txt'
        if os.path.exists(fp):
            accessOk = False
            for i in range(3):
                try:
                    with open(fp) as f:
                        dataList = f.readlines()
                    if do_reset:
                        os.remove(fp)
                    else:
                        for line in dataList:
                            (u, t) = line.split(':')
                            (t, m) = t.split('-')
                            (tin, tout) = t.split('/')
                            incount = int(tin)
                            outcount = int(tout)
                            if 'mini' in m:
                                tokens4mini_in += incount
                                tokens4mini_out += outcount
                                ttotal4mini_in += incount
                                ttotal4mini_out += outcount
                            else:
                                tokens4o_in += incount
                                tokens4o_out += outcount
                                ttotal4o_in += incount
                                ttotal4o_out += outcount
                    accessOk = True
                    break
                except:
                    sleep(3)
            if not accessOk:
                return f'File access failed reading stats for user: {user}'
        userAudio = 0
        fp = dataDir + user + '_audio.txt'
        if os.path.exists(fp):
            accessOk = False
            for i in range(3):
                try:
                    with open(fp) as f:
                        dataList = f.readlines()
                    if do_reset:
                        os.remove(fp)
                    else:
                        for line in dataList:
                            (dud, len) = line.split(':')
                            userAudio += int(len)
                        totalAudio += int(userAudio)
                    accessOk = True
                    break
                except:
                    sleep(3)
            if not accessOk:
                return f'File access failed reading audio stats for user: {user}'
        result.append([user, f'{tokens4mini_in}/{tokens4mini_out}', f'{tokens4o_in}/{tokens4o_out}', f'audio:{userAudio}'])
    result.append(['totals', f'{ttotal4mini_in}/{ttotal4mini_out}', f'{ttotal4o_in}/{ttotal4o_out}', f'audio:{totalAudio}'])
    return result       

def clear():
    return [None, [], None]

def updatePassword(txt):
    return [txt.lower().strip(), "*********"]

# def setModel(val):
#     return val

def chat(prompt, user_window, pwd_window, past, response, gptModel):
    user_window = user_window.lower().strip()
    isBoss = False
    if user_window == unames[0] and pwd_window == pwdList[0]:
        isBoss = True
        if prompt == 'stats':
            response = genUsageStats()
            return [past, response, None, gptModel]
        if prompt == 'reset':
            response = genUsageStats(True)
            return [past, response, None, gptModel]
        if prompt.startswith('gpt4'):
            gptModel = 'gpt-4o'
            prompt = prompt[5:]
    if user_window in unames and pwd_window in pwdList:
        past.append({"role":"user", "content":prompt})
        completion = client.chat.completions.create(model=gptModel,
                                            messages=past)
        reply = completion.choices[0].message.content
        tokens_in = completion.usage.prompt_tokens
        tokens_out = completion.usage.completion_tokens
        tokens = completion.usage.total_tokens
        response += "\n\nYOU: " + prompt + "\nGPT: " + reply
        if isBoss:
            response += f"\n{gptModel}: tokens in/out = {tokens_in}/{tokens_out}"
        if tokens > 40000:
            response += "\n\nTHIS DIALOG IS GETTING TOO LONG. PLEASE RESTART CONVERSATION SOON."
        past.append({"role":"assistant", "content": reply})
        accessOk = False
        for i in range(3):
            try:
                dataFile = new_func(user_window)
                with open(dataFile, 'a') as f:
                    m = '4o'
                    if 'mini' in gptModel:
                        m = '4omini'
                    f.write(f'{user_window}:{tokens_in}/{tokens_out}-{m}\n')
                accessOk = True
                break
            except Exception as e:
                sleep(3)
        if not accessOk:
            response += f"\nDATA LOG FAILED, path = {dataFile}"
        return [past, response , None, gptModel]
    else:
        return [[], "User name and/or password are incorrect", prompt, gptModel]

def new_func(user_window):
    dataFile = dataDir + user_window + '_log.txt'
    return dataFile

def play_sound(txt):
    if len(txt) < 10:
        gr.Info(message='Dialog must be at least 10 characters long',duration=4)
        return
    speech_path = dataDir + 'speech.wav'
    response = client.audio.speech.create(model='tts-1', voice='fable', input=txt)
    with open(speech_path, 'wb') as fpath:
        fpath.write(response.content)
    playsound(speech_path)
    if os.path.exists(speech_path):
        os.remove(speech_path)


def transcribe(user, pwd, fpath):
    user = user.lower().strip()
    pwd = pwd.lower().strip()
    if not (user in unames and pwd in pwdList):
        return 'Bad credentials'
    with audioread.audio_open(fpath) as audio:
        duration = int(audio.duration)
        if duration > 0:
            with open(dataDir + user + '_audio.txt','a') as f:
                f.write(f'audio:{str(duration)}\n')
    with open(fpath,'rb') as audio_file:
        transcript = client.audio.transcriptions.create(
            model='whisper-1', file = audio_file ,response_format = 'text' )
    reply = transcript
    return str(reply)

def pause_message():
    return "Audio input is paused.  Resume or Stop as desired"

def set_speak(txt):
    vis = False;
    if len(txt) > 10:
        vis = True
    return gr.Button(visible=vis)


with gr.Blocks() as demo:
    history = gr.State([])
    password = gr.State("")
    model = gr.State("gpt-4o-mini")
    gr.Markdown('# GPT Chat')
    gr.Markdown('Enter user name & password then enter prompt and click submit button.  Restart conversation if topic changes')
    gr.Markdown('You can enter prompts by voice.  Tap Record, speak, then tap Stop.' +
              '  Tap "Reset Voice Entry", to enter more voice.  Note: first voice response takes a long time.')
    # heading = gr.Label(value="GPT Chat", scale=2, color="Crimson" )
    with gr.Row():
        user_window = gr.Textbox(label = "User Name")
        pwd_window = gr.Textbox(label = "Password")
        pwd_window.blur(updatePassword, pwd_window, [password, pwd_window])
    with gr.Row():
        audio_widget = gr.Audio(type='filepath', format='wav',waveform_options=gr.WaveformOptions(
           show_recording_waveform=True), sources=['microphone'], scale = 3, label="Prompt/Question Voice Entry", max_length=120)
        reset_button = gr.ClearButton(value="Reset Voice Entry", scale=1) #new_func1()
    with gr.Row():
        clear_button = gr.Button(value="Restart Conversation")
        # gpt_chooser=gr.Radio(choices=[("GPT-3.5","gpt-3.5-turbo"),("GPT-4o","gpt-4o-mini")],
        #                      value="gpt-3.5-turbo", label="GPT Model", interactive=True)
        submit_window = gr.Button(value="Submit Prompt/Question")
        speak_dialog = gr.Button(value="Speak Dialog",visible=False)
    prompt_window = gr.Textbox(label = "Prompt or Question")
    output_window = gr.Textbox(label = "Dialog")
    submit_window.click(chat, inputs=[prompt_window, user_window, password, history, output_window, model],
                         outputs=[history, output_window, prompt_window, model])
    clear_button.click(clear, inputs=[], outputs=[prompt_window, history, output_window])
    audio_widget.stop_recording(fn=transcribe, inputs=[user_window, password, audio_widget],
                                outputs=[prompt_window])
    audio_widget.pause_recording(fn=pause_message, outputs=[prompt_window])
    reset_button.add(audio_widget)
    speak_dialog.click(play_sound, output_window, None)
    output_window.change(set_speak, output_window, speak_dialog)
demo.queue().launch(share=True)