app.py

from transformers import pipeline
from langchain.vectorstores import FAISS
from langchain.embeddings import HuggingFaceEmbeddings

pipe = pipeline("automatic-speech-recognition", model="openai/whisper-large-v2")
import json
with open("tasks.json", "r",encoding="utf-8") as json_file:
  global data
  data = json.load(json_file)
def find_index(sentence):
  global data
  for key, value in data.items():
    for i,j in value.items():
      for s in j:
        if sentence == s:
          return i
for x,item in data.items():
    texts = []
    for key,value in item.items():
      for each  in value:
        print(find_index(each))
        texts.append(each)
    globals()[f"faiss_{x}"] = FAISS.from_texts(texts,HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2",model_kwargs={'device':'cpu'}))


def transcribe_the_command(audio_path,state):
    transcript = pipe(audio_path)["text"]
    similar = globals()[f"faiss_{state}"].similarity_search(transcript)[0].page_content
    print(similar)
    reply = find_index(similar)
    return reply
import gradio as gr
iface = gr.Interface(
    fn=transcribe_the_command,
    inputs=[gr.Audio(source="microphone"),gr.Textbox()],
    outputs="text",
    title="Whisper Small",
    description="Realtime demo for intent recognition using a Whisper small model.",
)

iface.launch(share="true")