app.py

import gradio as gr
import torch
import torchaudio
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import librosa
import numpy as np
import re

rocessor = Wav2Vec2Processor.from_pretrained("adilism/wav2vec2-large-xlsr-kyrgyz")
model = Wav2Vec2ForCTC.from_pretrained("adilism/wav2vec2-large-xlsr-kyrgyz")
# model.to("cuda")

def transcribe(file_):
    arr_audio, _ librosa.load(file, rate=16000)
    inputs = processor(arr_audio, sampling_rate=16_000, return_tensors="pt", padding=True)

    with torch.no_grad():
        logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits

    pred_ids = torch.argmax(logits, dim=-1)
    text = processor.batch_decode(pred_ids)
    return text




iface = gr.Interface(
    fn=transcribe, 
    inputs=gr.Audio(type="filepath"),
    outputs="text",
    title="Kyrgyz-STT-Small",
    description="Realtime demo for Kyrgyz speech recognition using a fine-tuned Whisper small model.",
)

iface.launch()