app.py

import gradio as gr
import soundfile as sf
import torch
import numpy as np
import librosa
from transformers import AutoProcessor, Wav2Vec2BertForCTC
import spaces

MODEL_NAME = "mikr/w2v-bert-2.0-czech-colab-cv16"

device = 0 if torch.cuda.is_available() else "cpu"

print("device:",device)

processor = AutoProcessor.from_pretrained(MODEL_NAME)
model = Wav2Vec2BertForCTC.from_pretrained(MODEL_NAME).to(device)


@spaces.GPU
def transcribe(audio_path):
    a, s = librosa.load(audio_path, sr=16_000)

    # inputs = processor(a, sampling_rate=s, return_tensors="pt")
    input_values = processor(a, sampling_rate=s, return_tensors="pt").input_features

    with torch.no_grad():
        logits = model(input_values.to(device)).logits
   
    predicted_ids = torch.argmax(logits, dim=-1)

    # transcribe speech
    transcription = processor.batch_decode(predicted_ids)
    return transcription[0]


iface = gr.Interface(
    fn=transcribe,
    inputs=[
        gr.Audio(sources="upload", type="filepath", label="Upload Audio File"),  # Audio file upload
    ],
    outputs="text",
    theme="huggingface",
    title="Czech W2V-BERT 2.0 speech encoder demo - transcribe Czech Audio",
    description=(
        "Transcribe audio inputs with the click of a button! Demo uses the fine-tuned"
        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) from Facebook W2V-BERT 2.0 speech encoder "
        "and 🤗 Transformers to transcribe audio files of arbitrary length."
    ),
    allow_flagging="never",
)

iface.launch(server_name="0.0.0.0")