import gradio as gr
import torch
from transformers import WhisperForConditionalGeneration, WhisperProcessor, WhisperTokenizer
import librosa

model_name = "shReYas0363/whisper-tiny-fine-tuned" 
processor = WhisperProcessor.from_pretrained("openai/whisper-tiny")
tokenizer=WhisperTokenizer.from_pretrained("openai/whisper-tiny")
model = WhisperForConditionalGeneration.from_pretrained(model_name)

def transcribe(audio_file_path):
    audio, sr = librosa.load(audio_file_path, sr=16000)
    input_features = processor(audio, return_tensors="pt", sampling_rate=16000).input_features
    with torch.no_grad():
        predicted_ids = model.generate(input_features)
    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
    return transcription

interface = gr.Interface(
    fn=transcribe,
    inputs=gr.Audio(sources=["upload"], type="filepath"),
    outputs=gr.Textbox(label="Transcription"),
    title="Whisper Tiny ASR",
    description="Fine-tuned on AI4Bharat's svarah dataset"
)

interface.launch()