app.py

import accelerate
import gradio as gr
import time
import io
import librosa
import torch
import soundfile as sf

from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

#Instantiating the model object. 

model = AutoModelForSpeechSeq2Seq.from_pretrained(pretrained_model_name_or_path= "openai/whisper-large-v3",
                                                  torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
                                                  use_safetensors=True).to("cuda")

#Instantiating the processor object.

processor = AutoProcessor.from_pretrained(pretrained_model_name_or_path="openai/whisper-large-v3")

#Instantiating the transformer class' pipeline object.

pipe = pipeline(task="automatic-speech-recognition",
                model="openai/whisper-large-v3",
                tokenizer=processor.tokenizer,
                feature_extractor=processor.feature_extractor,
                max_new_tokens=128,
                chunk_length_s=30,
                batch_size=16,
                return_timestamps=True,
                torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
                device="cuda")

#Defining speech-to-text function.

def convert(audio, state=""):
    """
    This function performs speech to text conversion and will be used in Gradio's Interface function.
    Parameters:
    - audio: audio data as a bytes-like object.
    - state: a string representing the accumulated text from previous conversions.
    """
    time.sleep(3)
    try:
        result = pipe(audio)
        transcribed_text = result['text']
        state += transcribed_text + " "
    except Exception as e:
        return f"Error processing audio: Please start recording!", state

    return state, state

#Instantiating Gradio Interface.

gr_interface = gr.Interface(
    fn = convert,
    title = "Automatic Speech-to-Text",
    description = "### Record your speech and watch it get converted to text!",
    inputs = [
        gr.Audio(
            label="Please Record Your Speech Here!",
            sources="microphone",
            type="filepath"),
        "state"],
    outputs = [
        "textbox",
        "state"
    ],
    theme="dark",
    live=True
)

#Launching the app (share=True).

gr_interface.launch()