Create app.py

app.py

@@ -0,0 +1,140 @@
+import gradio as gr
+import torch
+import librosa
+from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq, pipeline
+
+# -------------------------------
+# 1) STT 모델 로드 (Whisper 기반)
+# -------------------------------
+processor = AutoProcessor.from_pretrained("openai/whisper-small")
+model = AutoModelForSpeechSeq2Seq.from_pretrained("openai/whisper-small")
+
+# -------------------------------
+# 2) 감정 분석 모델 로드
+# -------------------------------
+sentiment_pipe = pipeline(
+    "sentiment-analysis",
+    model="monologg/koelectra-base-v3-discriminator",
+    tokenizer="monologg/koelectra-base-v3-discriminator"
+)
+
+# -------------------------------
+# 3) 오디오 -> 텍스트
+# -------------------------------
+def transcribe_audio(audio_path):
+    speech, _ = librosa.load(audio_path, sr=16000)
+    input_features = processor(speech, sampling_rate=16000, return_tensors="pt").input_features
+    predicted_ids = model.generate(input_features)
+    transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
+    return transcription
+
+# -------------------------------
+# 4) 감정 레이블 -> 색상 매핑
+# -------------------------------
+def label_to_color(label):
+    if label in ["4 stars", "5 stars", "LABEL_4", "LABEL_5"]:
+        return "green"  # 긍정
+    elif label in ["1 star", "2 stars", "LABEL_1", "LABEL_2"]:
+        return "red"    # 부정
+    else:  # 중립
+        return "orange"
+
+# -------------------------------
+# 5) 텍스트 감정 분석 (문장 전체)
+# -------------------------------
+def sentiment_whole_text(text):
+    res = sentiment_pipe(text)[0]
+    label = res['label']
+    score = res['score']
+    color = label_to_color(label)
+    styled_text = f"<span style='color:{color}'>{text}</span>"
+    legend = (
+        "<div style='margin-top:10px;'>"
+        "<b>색상 설명:</b> "
+        "<span style='color:green'>녹색=긍정</span>, "
+        "<span style='color:red'>빨강=부정</span>, "
+        "<span style='color:orange'>주황=중립/보통</span>"
+        "</div>"
+    )
+    return styled_text + legend, f"감정: {label}, 신뢰도: {score:.2f}"
+
+# -------------------------------
+# 6) 텍스트 감정 분석 (단어별)
+# -------------------------------
+def sentiment_word_level(text):
+    words = text.split()
+    styled_words = []
+    for w in words:
+        res = sentiment_pipe(w)[0]
+        label = res['label']
+        color = label_to_color(label)
+        styled_words.append(f"<span style='color:{color}'>{w}</span>")
+    styled_text = " ".join(styled_words)
+    legend = (
+        "<div style='margin-top:10px;'>"
+        "<b>색상 설명:</b> "
+        "<span style='color:green'>녹색=긍정</span>, "
+        "<span style='color:red'>빨강=부정</span>, "
+        "<span style='color:orange'>주황=중립/보통</span>"
+        "</div>"
+    )
+    return styled_text + legend, "단어별 감정 표시 완료"
+
+# -------------------------------
+# 7) 오디오 -> 텍스트 + 감정 분석 (문장+단어)
+# -------------------------------
+def process_audio_full(audio_file):
+    text = transcribe_audio(audio_file)
+    whole_text_result, whole_text_score = sentiment_whole_text(text)
+    word_level_result, word_level_status = sentiment_word_level(text)
+    return text, whole_text_result, whole_text_score, word_level_result, word_level_status
+
+# -------------------------------
+# 8) Gradio UI 구성
+# -------------------------------
+with gr.Blocks() as demo:
+    gr.Markdown("# 🎤 오디오/텍스트 → 감정 분석")
+    
+    with gr.Tabs():
+        # ------------------- 오디오 -> 텍스트 -------------------
+        with gr.Tab("오디오 → 텍스트"):
+            audio_input_1 = gr.Audio(label="음성 업로드", type="filepath")
+            audio_text_output = gr.Textbox(label="변환된 텍스트")
+            audio_transcribe_btn = gr.Button("텍스트 추출")
+            audio_transcribe_btn.click(fn=transcribe_audio, inputs=[audio_input_1], outputs=[audio_text_output])
+        
+        # ------------------- 텍스트 -> 감정 분석 -------------------
+        with gr.Tab("텍스트 → 감정 분석"):
+            text_input = gr.Textbox(label="텍스트 입력")
+            sentiment_whole_output = gr.HTML(label="문장 단위 감정 분석")
+            sentiment_whole_score = gr.Markdown(label="감정 결과")
+            sentiment_word_output = gr.HTML(label="단어 단위 감정 분석")
+            sentiment_btn = gr.Button("감정 분석")
+            def analyze_text(text):
+                whole_res, whole_score = sentiment_whole_text(text)
+                word_res, word_status = sentiment_word_level(text)
+                return whole_res, whole_score, word_res
+            sentiment_btn.click(
+                fn=analyze_text,
+                inputs=[text_input],
+                outputs=[sentiment_whole_output, sentiment_whole_score, sentiment_word_output]
+            )
+        
+        # ------------------- 오디오 → 텍스트 + 감정 분석 -------------------
+        with gr.Tab("오디오 → 텍스트 + 감정 분석"):
+            audio_input_2 = gr.Audio(label="음성 업로드", type="filepath")
+            audio_text_output_2 = gr.Textbox(label="변환된 텍스트")
+            sentiment_whole_output_2 = gr.HTML(label="문장 단위 감정 분석")
+            sentiment_whole_score_2 = gr.Markdown(label="감정 결과")
+            sentiment_word_output_2 = gr.HTML(label="단어 단위 감정 분석")
+            audio_process_btn = gr.Button("분석 시작")
+            def process_audio_tab(audio_file):
+                text, whole_res, whole_score, word_res, word_status = process_audio_full(audio_file)
+                return text, whole_res, whole_score, word_res
+            audio_process_btn.click(
+                fn=process_audio_tab,
+                inputs=[audio_input_2],
+                outputs=[audio_text_output_2, sentiment_whole_output_2, sentiment_whole_score_2, sentiment_word_output_2]
+            )
+
+demo.launch()