app.py

from transformers import pipeline,WhisperProcessor
import gradio as gr
import os

# Set the cache directory for Hugging Face models to ensure they are saved within the Space
os.environ['HUGGINGFACE_HUB_CACHE'] = '/app/.cache/huggingface/hub'

# The name of the model you want to use
model_name = "16pramodh/ASR_YAP"
processor = WhisperProcessor.from_pretrained("openai/whisper-small.en")

# Load the ASR pipeline
pipe = pipeline(
    "automatic-speech-recognition",
    model=model_name,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    device=0, # Use GPU if available on paid tiers, otherwise falls back to CPU
)

# Define the transcription function that Gradio will expose as an API
def transcribe_audio(audio_file_path):
    if audio_file_path is None:
        return "No audio file provided."

    # The pipeline can directly process the path to the audio file
    transcription = pipe(audio_file_path)["text"]
    return transcription

# Create the Gradio Interface
iface = gr.Interface(
    fn=transcribe_audio,
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="Whisper Indian Accent ASR API",
    description="An API endpoint for a fine-tuned Whisper model.",
)

# Launch the Gradio app
if __name__ == "__main__":
    iface.launch()