app.py

import gradio as gr
from transformers import pipeline
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import json
import os
import torch

data_dict = {}
with open('./intents.json', 'r') as file:
    data = json.load(file)    
intents_dict = data
tokenizer = AutoTokenizer.from_pretrained("roberta-base")
model = AutoModelForSequenceClassification.from_pretrained("./")

def preprocess(text):
    inputs = tokenizer(text, padding=True, truncation=True, return_tensors="pt")
    return inputs

def postprocess(outputs):
    logits = outputs.logits
    predicted_labels = logits.argmax(dim=1).tolist()
    return predicted_labels

def predict(text):
    inputs = preprocess(text)
    with torch.no_grad():
        outputs = model(**inputs)
    predicted_labels = postprocess(outputs)
    ans = intents_dict[predicted_labels[0]]
    return ans

from transformers import pipeline

p = pipeline(model="openai/whisper-medium")

def transcribe(audio):
    if audio["filepath"]:
        audio["text"] = p(audio["filepath"])['text']
    ans = predict(audio["text"])
    return ans


get_intent = gr.Interface(fn = transcribe,
                          inputs=[gr.Textbox(label="Enter Text Input", type="text"),gr.Audio(source="microphone", type="filepath")],
                          outputs="text").launch()