|
import gradio as gr |
|
import numpy as np |
|
import librosa |
|
from transformers import pipeline |
|
from datetime import datetime |
|
import os |
|
import requests |
|
|
|
|
|
API_URL = "https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-xl-base-1.0" |
|
headers = {"Authorization": "Bearer hf_..."} |
|
|
|
|
|
speech_recognizer = pipeline( |
|
"automatic-speech-recognition", |
|
model="kresnik/wav2vec2-large-xlsr-korean" |
|
) |
|
emotion_classifier = pipeline( |
|
"audio-classification", |
|
model="MIT/ast-finetuned-speech-commands-v2" |
|
) |
|
text_analyzer = pipeline( |
|
"sentiment-analysis", |
|
model="nlptown/bert-base-multilingual-uncased-sentiment" |
|
) |
|
|
|
def create_interface(): |
|
with gr.Blocks(theme=gr.themes.Soft()) as app: |
|
state = gr.State({ |
|
"user_name": "", |
|
"reflections": [], |
|
"voice_analysis": None, |
|
"final_prompt": "" |
|
}) |
|
def generate_image_from_prompt(prompt): |
|
"""HuggingFace Inference API를 통한 이미지 생성""" |
|
try: |
|
response = requests.post(API_URL, headers=headers, json={ |
|
"inputs": prompt, |
|
"parameters": { |
|
"negative_prompt": "ugly, blurry, poor quality, distorted", |
|
"num_inference_steps": 30, |
|
"guidance_scale": 7.5 |
|
} |
|
}) |
|
|
|
if response.status_code == 200: |
|
return response.content |
|
else: |
|
return None |
|
except Exception as e: |
|
print(f"Error generating image: {e}") |
|
return None |
|
|
|
|
|
|
|
|
|
header = gr.Markdown("# 디지털 굿판") |
|
user_display = gr.Markdown("") |
|
|
|
with gr.Tabs() as tabs: |
|
|
|
with gr.Tab("입장"): |
|
gr.Markdown("""# 디지털 굿판에 오신 것을 환영합니다""") |
|
name_input = gr.Textbox(label="이름을 알려주세요") |
|
start_btn = gr.Button("여정 시작하기") |
|
|
|
|
|
with gr.Tab("청신"): |
|
with gr.Row(): |
|
|
|
audio_path = os.path.abspath(os.path.join("assets", "main_music.mp3")) |
|
audio = gr.Audio( |
|
value=audio_path, |
|
type="filepath", |
|
label="온천천의 소리", |
|
interactive=False, |
|
autoplay=True |
|
) |
|
with gr.Column(): |
|
reflection_input = gr.Textbox( |
|
label="현재 순간의 감상을 적어주세요", |
|
lines=3 |
|
) |
|
save_btn = gr.Button("감상 저장하기") |
|
reflections_display = gr.Dataframe( |
|
headers=["시간", "감상", "감정 분석"], |
|
label="기록된 감상들" |
|
) |
|
|
|
|
|
with gr.Tab("기원"): |
|
gr.Markdown("## 기원 - 목소리로 전하기") |
|
with gr.Row(): |
|
with gr.Column(): |
|
voice_input = gr.Audio( |
|
label="나누고 싶은 이야기를 들려주세요", |
|
sources=["microphone"], |
|
type="filepath", |
|
interactive=True |
|
) |
|
clear_btn = gr.Button("녹음 지우기") |
|
|
|
with gr.Column(): |
|
transcribed_text = gr.Textbox( |
|
label="인식된 텍스트", |
|
interactive=False |
|
) |
|
voice_emotion = gr.Textbox( |
|
label="음성 감정 분석", |
|
interactive=False |
|
) |
|
text_emotion = gr.Textbox( |
|
label="텍스트 감정 분석", |
|
interactive=False |
|
) |
|
analyze_btn = gr.Button("분석하기") |
|
|
|
|
|
generate_btn.click( |
|
fn=generate_image_from_prompt, |
|
inputs=[final_prompt], |
|
outputs=[result_image] |
|
) |
|
|
|
with gr.Tab("송신"): |
|
gr.Markdown("## 송신 - 시각화 결과") |
|
with gr.Column(): |
|
final_prompt = gr.Textbox( |
|
label="생성된 프롬프트", |
|
interactive=False, |
|
lines=3 |
|
) |
|
generate_btn = gr.Button("이미지 생성하기") |
|
result_image = gr.Image( |
|
label="생성된 이미지", |
|
type="pil" |
|
) |
|
|
|
def clear_voice_input(): |
|
"""음성 입력 초기화""" |
|
return None |
|
|
|
def analyze_voice(audio_path, state): |
|
"""음성 분석""" |
|
if audio_path is None: |
|
return state, "음성을 먼저 녹음해주세요.", "", "", "" |
|
|
|
try: |
|
|
|
y, sr = librosa.load(audio_path, sr=16000) |
|
|
|
|
|
transcription = speech_recognizer(y) |
|
text = transcription["text"] |
|
|
|
|
|
voice_emotions = emotion_classifier(y) |
|
text_sentiment = text_analyzer(text)[0] |
|
|
|
|
|
prompt = generate_prompt(text, voice_emotions[0], text_sentiment) |
|
|
|
return ( |
|
state, |
|
text, |
|
f"음성 감정: {voice_emotions[0]['label']} ({voice_emotions[0]['score']:.2f})", |
|
f"텍스트 감정: {text_sentiment['label']} ({text_sentiment['score']:.2f})", |
|
prompt |
|
) |
|
except Exception as e: |
|
return state, f"오류 발생: {str(e)}", "", "", "" |
|
|
|
def generate_prompt(text, voice_emotion, text_sentiment): |
|
"""프롬프트 생성""" |
|
emotion_colors = { |
|
"happy": "따뜻한 노란색과 주황색", |
|
"sad": "깊은 파랑색과 보라색", |
|
"angry": "강렬한 빨강색과 검정색", |
|
"neutral": "부드러운 회색과 베이지색" |
|
} |
|
|
|
color = emotion_colors.get(voice_emotion['label'], "자연스러운 색상") |
|
prompt = f"한국 전통 민화 스타일의 추상화, {color} 사용. " |
|
prompt += f"음성의 감정({voice_emotion['label']})과 텍스트의 감정({text_sentiment['label']})이 조화를 이루며, " |
|
prompt += f"음성의 특징을 반영한 동적인 구도. 발화 내용: '{text}'" |
|
|
|
return prompt |
|
|
|
def save_reflection(text, state): |
|
"""감상 저장""" |
|
if not text.strip(): |
|
return state, state["reflections"] |
|
|
|
current_time = datetime.now().strftime("%H:%M:%S") |
|
sentiment = text_analyzer(text)[0] |
|
new_reflection = [current_time, text, f"{sentiment['label']} ({sentiment['score']:.2f})"] |
|
|
|
if "reflections" not in state: |
|
state["reflections"] = [] |
|
|
|
state["reflections"].append(new_reflection) |
|
return state, state["reflections"] |
|
|
|
|
|
start_btn.click( |
|
fn=lambda name: (f"# 환영합니다, {name}님의 디지털 굿판", gr.update(selected="청신")), |
|
inputs=[name_input], |
|
outputs=[user_display, tabs] |
|
) |
|
|
|
save_btn.click( |
|
fn=save_reflection, |
|
inputs=[reflection_input, state], |
|
outputs=[state, reflections_display] |
|
) |
|
|
|
clear_btn.click( |
|
fn=clear_voice_input, |
|
inputs=[], |
|
outputs=[voice_input] |
|
) |
|
|
|
analyze_btn.click( |
|
fn=analyze_voice, |
|
inputs=[voice_input, state], |
|
outputs=[state, transcribed_text, voice_emotion, text_emotion, final_prompt] |
|
) |
|
|
|
return app |
|
|
|
if __name__ == "__main__": |
|
demo = create_interface() |
|
demo.launch() |