app.py

import gradio as gr
import torch
from transformers import AutoProcessor, AutoModelForCTC
import soundfile as sf  # For handling audio input

# Load model directly
from transformers import AutoTokenizer, AutoModelForPreTraining
tokenizer = AutoTokenizer.from_pretrained("Opit/mms_tts_bulgarian_finetuning")
model = AutoModelForPreTraining.from_pretrained("Opit/mms_tts_bulgarian_finetuning")

# ASR 변환 함수 (speech-to-text conversion)
def asr_generate(audio):
    # Load and process the audio file
    speech, _ = sf.read(audio)
    inputs = processor(speech, sampling_rate=16000, return_tensors="pt", padding=True)

    with torch.no_grad():
        logits = model(**inputs).logits

    # Get predicted IDs and decode the text
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.batch_decode(predicted_ids)[0]
    
    return transcription


# Gradio 인터페이스 생성
iface = gr.Interface(
    fn=asr_generate, 
    inputs=gr.Audio(source="microphone", type="filepath"), 
    outputs="text",
    title="Bulgarian Speech Recognition",
    description="Upload or record audio in Bulgarian to get the transcription."
)

# 인터페이스 실행
if __name__ == "__main__":
    iface.launch()