app.py

# Gaepago model V1 (CPU Test)

# import package
from transformers import AutoModelForAudioClassification
from transformers import AutoFeatureExtractor
from transformers import pipeline
from datasets import Dataset, Audio
import gradio as gr
import torch

# Set model & Dataset NM
MODEL_NAME = "Gae8J/gaepago-20"
DATASET_NAME = "Gae8J/modeling_v1"

# Import Model & feature extractor
# model = AutoModelForAudioClassification.from_pretrained(MODEL_NAME)
from transformers import AutoConfig
config = AutoConfig.from_pretrained(MODEL_NAME)
model = torch.jit.load(f"./model/gaepago-20-lite/model_quant_int8.pt")
feature_extractor = AutoFeatureExtractor.from_pretrained(MODEL_NAME)

# 모델 cpu로 변경하여 진행
model.to("cpu")

# Gaepago Inference Model function
def gaepago_fn(tmp_audio_dir):
    print(tmp_audio_dir)
    audio_dataset = Dataset.from_dict({"audio": [tmp_audio_dir]}).cast_column("audio", Audio(sampling_rate=16000))
    inputs = feature_extractor(audio_dataset[0]["audio"]["array"]
                               ,sampling_rate=audio_dataset[0]["audio"]["sampling_rate"]
                               ,return_tensors="pt")
    with torch.no_grad():
        # logits = model(**inputs).logits
        logits = model(**inputs)["logits"]        
    # predicted_class_ids = torch.argmax(logits).item()
    # predicted_label = model.config.id2label[predicted_class_ids]
    predicted_class_ids = torch.argmax(logits).item()
    predicted_label = config.id2label[predicted_class_ids]
    
    return predicted_label

# Main
main_api = gr.Blocks()

with main_api:
    gr.Markdown("## 8J Gaepago Demo(with CPU)")
    with gr.Row():
        audio = gr.Audio(source="microphone", type="filepath"
                         ,label='녹음버튼을 눌러 초코가 하는 말을 들려주세요')
        transcription = gr.Textbox(label='지금 초코가 하는 말은...')
    b1 = gr.Button("강아지 언어 번역!")

    b1.click(gaepago_fn, inputs=audio, outputs=transcription)
#    examples = gr.Examples(examples=example_list,
#                           inputs=[audio])
main_api.launch()