app.py

import torch
from transformers import pipeline
import json
import requests
from io import BytesIO

transcribe = pipeline(task = "automatic-speech-recognition", model = "vasista22/whisper-tamil-medium", chunk_length_s=30, device="cpu")
transcribe.model.config.forced_decoder_ids = transcribe.tokenizer.get_decoder_prompt_ids(language="ta", task="transcribe")

def process_audio(params):
    # DEEPGRAM_API_KEY = "57642b9249bc230f8055c4daeeb0a866d28f6491"
    solutions=[]
    file_ids = []

    params = json.loads(params)

    audio_files = params.get("urls",[])
    if not params.get("normalfileID",[]):
        file_ids = [None]*len(audio_files)
    else:
        file_ids = params.get("normalfileID",[])
    print(params)
    for audio, file_id in zip(audio_files,file_ids):
        # deepgram = DeepgramClient(DEEPGRAM_API_KEY)

        # options = PrerecordedOptions(
        #     model="enhanced",
        #     language="ta",
        # )
        print("API hit")

        # response = deepgram.listen.prerecorded.v("1").transcribe_url({"url": audio}, options)

        # text = response['results']['channels'][0]['alternatives'][0]['transcript']
        print(audio)
        response = requests.get(audio)
        audio2 = BytesIO(response.content)
        text = transcribe(audio2)["text"]

        answer_dict = {}
        answer_dict.update({'audioURL':audio, 'solutions':text, "qcUserId": None, "normalfileID": file_id})
        solutions.append(answer_dict)

    # result_url = f"{api}/{job_id}"
    # send_results_to_api(solutions, result_url)
    return json.dumps({"solutions":solutions})

import gradio as gr
inputt = gr.Textbox(label = "Parameter in json format   Eg. {'audio_files':['file1.mp3','file2.wav'], 'api':'https://api.example.com', 'job_id':'1001'}")
outputt = gr.JSON()

application = gr.Interface(fn=process_audio, inputs = inputt, outputs= outputt, title="Multilingual (Hindi/English) Audio transcription with API Intergration")
application.launch()