app.py

"""
this model only supports english since text to speech is an english only model
"""
from google.cloud import texttospeech
import os
import openai
import gradio as gr
from dotenv import load_dotenv
import pinecone


"""
login to gcp
"""
os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = "gcp_access_key.json"
# Instantiates a client
client = texttospeech.TextToSpeechClient()

""" 
Connecting to Open AI API
"""
load_dotenv()
openai.organization = os.getenv("OPENAI_ORG")
openai.api_key = os.getenv("OPENAI_API_KEY")
EMBEDDING_MODEL = "text-embedding-ada-002"
"""
Connecting to pincone API and assign index
"""
index_name = 'economic-forecast'
pinecone.init(
    api_key=os.getenv("Pinecone_KEY"),
    environment=os.getenv("Pinecone_ENV")
)

## initial a first message to define GPT's role


"""
define the text -> speech function
"""
def text2speech(text):

    # Set the text input to be synthesized
    synthesis_input = texttospeech.SynthesisInput(text=text)

    # Build the voice request, select the language code ("en-US") and the ssml
    # voice gender ("neutral")
    voice = texttospeech.VoiceSelectionParams(
        language_code="en-US", ssml_gender=texttospeech.SsmlVoiceGender.MALE
    )

    # Select the type of audio file you want returned
    audio_config = texttospeech.AudioConfig(
        audio_encoding=texttospeech.AudioEncoding.MP3
    )

    # Perform the text-to-speech request on the text input with the selected
    # voice parameters and audio file type
    response = client.synthesize_speech(
        input=synthesis_input, voice=voice, audio_config=audio_config
    )
    # The response's audio_content is binary.
    with open("output.mp3", "wb") as out:
        # Write the response to the output file.
        out.write(response.audio_content)
        print('Audio content written to file "output.mp3"')

"""
define voice -> gpt -> text -> voice workflow
"""
def transcribe(audio):
    #global messages

    """
    gradio output file doesn't have .wav so rename the file to the correct format
    """
    extension = ".wav"
    audiofomated = f"{audio}{extension}"
    os.rename(audio,audiofomated) 

    """
    pass the audio file to whisper to transcribe

    """
    audio_file = open(audiofomated, "rb")
    transcript = openai.Audio.transcribe("whisper-1", audio_file)

    
    """
    run cosin similarity to find context
    """
    ### Input the question and search for the relavent text
    index = pinecone.Index(index_name)
    query = openai.Embedding.create(input=transcript["text"], model=EMBEDDING_MODEL)["data"][0]["embedding"] # embed the user query into an embedding vector
    res = index.query(query, top_k=3, include_metadata=True) # run cosin similarity to search the most relavent embeded content; this is done in pinecone only
    contexts = [
            x['metadata']['text'] for x in res['matches']
        ]
    merged_context = "".join(contexts)
    contextwithQuestion = "Context: " + "\n"+ merged_context + "*End of the context*" + "\n\n" +  "Question: " + transcript["text"]


    """
    pass the transcripted text to GPT
    """
    messages = [
    {"role": "system", 
     "content": 
        "You are an assistant that answers questions only based on the context provided. Before each question, some context will be provided.\
        Context starts with 'Context:' and end with '*End of the context*'. Once you receive all the context, you will consider all of them to answer the questions.\
        It is very important to answer the question as honestly as possible.\
        If you are not sure about the answer based on the context provided, you can still try to come up with an answer but you must also tell the user that you are not confident about the answer and that the user should look for a secondary source to confirm the answer.\
        It is very important to answer the questions politely. It is very important to answer the question in great detail.\
        Once you receive all the context, you will receive a question that starts with 'Question:'. Once you receive the question, you can answer the question.\
        "}
] 
    messages.append({"role": "user", "content":contextwithQuestion}) ## add user input to the list of message
 
    response = openai.ChatCompletion.create(
        model="gpt-3.5-turbo",
        messages=messages
    ) ## pass the list of message to GPT

    messages.append({"role": "assistant", "content":response["choices"][0]["message"]["content"]}) ## add GPT response to the list of message
    text2speech(response["choices"][0]["message"]["content"]) ## create mp3 voice output
    
    voice_path = os.path.abspath("output.mp3")

    return voice_path


output_audio = gr.outputs.Audio(type = "filepath", label="AI Assistant")

gr.Interface(fn=transcribe, \
            inputs=gr.Audio(source="microphone", type="filepath",label="Speak here..."), \
            outputs=output_audio, \
            live=True,\
            allow_flagging='never')\
            .launch() ## add share=True to publish on the public site