Update app.py

app.py

@@ -1,126 +1,146 @@
 import gradio as gr
-import torch, librosa, zipfile, os, json, tempfile, uuid
+import torch
+import librosa
 from transformers import Wav2Vec2Processor, AutoModelForCTC
-from datetime import datetime, timedelta
+import zipfile
+import os
 import firebase_admin
 from firebase_admin import credentials, firestore, storage
+from datetime import datetime, timedelta
+import json
+tmpdir = None
+
+def transcribe(audio_file):
+    try:
+        audio, rate = librosa.load(audio_file, sr=16000)
+        input_values = processor(audio, sampling_rate=16000, return_tensors="pt").input_values
+        with torch.no_grad():
+            logits = model(input_values).logits
+        predicted_ids = torch.argmax(logits, dim=-1)
+        transcription = processor.batch_decode(predicted_ids)[0]
+        return transcription.replace("[UNK]", "")
+    except Exception as e:
+        return f"處理文件錯誤: {e}"
 
-# ---------- Firebase init ----------
-firebase_config = json.loads(os.environ.get("firebase_creds"))
+# Initialize Firebase
+firebase_config = json.loads(os.environ.get('firebase_creds'))
 cred = credentials.Certificate(firebase_config)
-firebase_admin.initialize_app(
-    cred, {"storageBucket": "amis-asr-corrections-dem-8cf3d.firebasestorage.app"}
-)
+firebase_admin.initialize_app(cred, {
+    "storageBucket": "amis-asr-corrections-dem-8cf3d.firebasestorage.app"
+})
 db = firestore.client()
 bucket = storage.bucket()
 
-# ---------- ASR model ----------
+# Load ASR model and processor
 MODEL_NAME = "eleferrand/XLSR_paiwan"
 processor = Wav2Vec2Processor.from_pretrained(MODEL_NAME)
 model = AutoModelForCTC.from_pretrained(MODEL_NAME)
 
-# ---------- Core helpers ----------
-def transcribe(path):
-    try:
-        audio, _ = librosa.load(path, sr=16_000)
-        inputs = processor(audio, sampling_rate=16_000, return_tensors="pt").input_values
-        with torch.no_grad():
-            logits = model(inputs).logits
-        ids = torch.argmax(logits, dim=-1)
-        text = processor.batch_decode(ids)[0]
-        return text.replace("[UNK]", "")
-    except Exception as e:
-        return f"處理文件錯誤: {e}"
-
-def transcribe_both(path):
-    txt = transcribe(path)
-    return txt, txt  # original & editable copies
+def transcribe_both(audio_file):
+    transcription = transcribe(audio_file)
+    return transcription, transcription
 
-def store_correction(orig, corr, audio, age, native):
+def store_correction(original_transcription, corrected_transcription, audio_file, age, native_speaker):
     try:
-        audio_meta, audio_url = {}, None
-        if audio and os.path.exists(audio):
-            a, sr = librosa.load(audio, sr=44_100)
-            audio_meta = {
-                "duration": librosa.get_duration(y=a, sr=sr),
-                "file_size": os.path.getsize(audio),
-            }
-            uid = f"{uuid.uuid4()}.wav"
-            blob = bucket.blob(f"audio/pai/{uid}")
-            blob.upload_from_filename(audio)
-            audio_url = blob.generate_signed_url(expiration=timedelta(hours=1))
-
-        db.collection("paiwan_transcriptions").add(
-            {
-                "transcription_info": {
-                    "original_text": orig,
-                    "corrected_text": corr,
-                    "language": "pai",
-                },
-                "audio_data": {"audio_metadata": audio_meta, "audio_file_url": audio_url},
-                "user_info": {"native_paiwan_speaker": native, "age": age},
-                "timestamp": datetime.now().isoformat(),
-                "model_name": MODEL_NAME,
-            }
-        )
-        return "校正保存成功! (Correction saved successfully!)"
+        audio_metadata = {}
+        audio_file_url = None
+        if audio_file and os.path.exists(audio_file):
+            audio, sr = librosa.load(audio_file, sr=44100)
+            duration = librosa.get_duration(y=audio, sr=sr)
+            file_size = os.path.getsize(audio_file)
+            audio_metadata = {'duration': duration, 'file_size': file_size}
+            unique_id = str(uuid.uuid4())
+            destination_path = f"audio/pai/{unique_id}.wav"
+            blob = bucket.blob(destination_path)
+            blob.upload_from_filename(audio_file)
+            audio_file_url = blob.generate_signed_url(expiration=timedelta(hours=1))
+        combined_data = {
+            'transcription_info': {'original_text': original_transcription, 'corrected_text': corrected_transcription, 'language': 'pai'},
+            'audio_data': {'audio_metadata': audio_metadata, 'audio_file_url': audio_file_url},
+            'user_info': {'native_paiwan_speaker': native_speaker, 'age': age},
+            'timestamp': datetime.now().isoformat(), 'model_name': MODEL_NAME
+        }
+        db.collection('paiwan_transcriptions').add(combined_data)
+        return "校正保存成功!"
     except Exception as e:
-        return f"保存失敗: {e} (Error saving correction: {e})"
+        return f"保存失败: {e}"
 
-def prepare_download(audio, orig, corr):
-    if not audio:
+def prepare_download(audio_file, original_transcription, corrected_transcription):
+    if audio_file is None:
         return None
     tmp_zip = tempfile.NamedTemporaryFile(delete=False, suffix=".zip")
     tmp_zip.close()
-    with zipfile.ZipFile(tmp_zip.name, "w") as z:
-        if os.path.exists(audio):
-            z.write(audio, arcname="audio.wav")
-        for name, txt in [("original_transcription.txt", orig),
-                          ("corrected_transcription.txt", corr)]:
-            with open(name, "w", encoding="utf-8") as f:
-                f.write(txt)
-            z.write(name, arcname=name)
-            os.remove(name)
+    with zipfile.ZipFile(tmp_zip.name, "w") as zf:
+        if os.path.exists(audio_file):
+            zf.write(audio_file, arcname="audio.wav")
+        orig_txt = "original_transcription.txt"
+        with open(orig_txt, "w", encoding="utf-8") as f:
+            f.write(original_transcription)
+        zf.write(orig_txt, arcname=orig_txt)
+        os.remove(orig_txt)
+        corr_txt = "corrected_transcription.txt"
+        with open(corr_txt, "w", encoding="utf-8") as f:
+            f.write(corrected_transcription)
+        zf.write(corr_txt, arcname=corr_txt)
+        os.remove(corr_txt)
     return tmp_zip.name
 
-# ---------- Interface ----------
+# Interface
 with gr.Blocks() as demo:
-    gr.Markdown("# 排灣語自動語音識別逐字稿與修正系統 (Paiwan ASR Transcription & Correction System)")
-
-    # Step 1
-    gr.Markdown("### 步驟 1：音訊上傳 (Audio Upload)")
-    gr.Markdown("上傳後請至步驟 2 按「產生逐字稿」，系統處理時請耐心等待…")
-    audio_input = gr.Audio(["upload", "microphone"], type="filepath",
-                           label="音訊輸入 (Audio Input)")
+    title = gr.Markdown("排灣語自動語音識別校正系統 (Paiwan ASR Transcription & Correction System)")
+    step1 = gr.Markdown(
+        "步驟 1：音訊上傳與產生逐字稿 (Audio Upload & Automatic Transcription)\n\n上傳後系統將自動產生逐字稿，請耐心等待。"
+    )
+    with gr.Row():
+        audio_input = gr.Audio(
+            sources=["upload", "microphone"], type="filepath", label="音訊輸入 (Audio Input)"
+        )
 
-    # Step 2
-    gr.Markdown("### 步驟 2：產生與編輯逐字稿 (Generate & Edit Transcript)")
-    trans_btn = gr.Button("產生逐字稿 (Generate Transcript)")
-    original = gr.Textbox(label="原始逐字稿 (Original Transcription)",
-                          interactive=False, lines=6)
-    corrected = gr.Textbox(label="更正逐字稿 (Corrected Transcription)",
-                           interactive=True, lines=6)
+    step2 = gr.Markdown("步驟 2：審閱與編輯逐字稿 (Step 2: Review & Edit Transcription)")
+    with gr.Row():
+        original_text = gr.Textbox(
+            label="原始逐字稿 (Original Transcription)", interactive=False, lines=5
+        )
+        corrected_text = gr.Textbox(
+            label="更正逐字稿 (Corrected Transcription)", interactive=True, lines=5
+        )
+    # Automatically generate transcription on audio upload
+    audio_input.change(
+        transcribe_both,
+        inputs=audio_input,
+        outputs=[original_text, corrected_text],
+        queue=True
+    )
 
-    # Step 3
-    gr.Markdown("### 步驟 3：使用者資訊 (User Information)")
+    step3 = gr.Markdown("步驟 3：使用者資訊 (Step 3: User Information)")
     with gr.Row():
-        age = gr.Slider(0, 100, step=1, value=25, label="年齡 (Age)")
-        native = gr.Checkbox(value=True, label="母語排灣語使用者？(Native Paiwan Speaker?)")
+        age_input = gr.Slider(
+            minimum=0, maximum=100, step=1, label="年齡 (Age)", value=25
+        )
+        native_speaker_input = gr.Checkbox(
+            label="母語排灣語使用者? (Native Paiwan Speaker?)", value=True
+        )
 
-    # Step 4
-    gr.Markdown("### 步驟 4：儲存與下載 (Save & Download)")
+    step4 = gr.Markdown("步驟 4：儲存與下載 (Step 4: Save & Download)")
     with gr.Row():
-        save_btn = gr.Button("儲存 (Save)")
-        save_msg = gr.Textbox(label="儲存狀態 (Save Status)", interactive=False)
+        save_button = gr.Button("儲存 (Save)")
+        save_status = gr.Textbox(
+            label="儲存狀態 (Save Status)", interactive=False
+        )
+
     with gr.Row():
-        dl_btn = gr.Button("下載 ZIP 檔案 (Download ZIP File)")
-        dl_out = gr.File()
+        download_button = gr.Button("下載 ZIP 檔案 (Download ZIP File)")
+        download_output = gr.File()
 
-    # --- wiring ---
-    trans_btn.click(transcribe_both, audio_input, [original, corrected])
-    save_btn.click(store_correction,
-                   [original, corrected, audio_input, age, native],
-                   save_msg)
-    dl_btn.click(prepare_download, [audio_input, original, corrected], dl_out)
+    save_button.click(
+        store_correction,
+        inputs=[original_text, corrected_text, audio_input, age_input, native_speaker_input],
+        outputs=save_status
+    )
+    download_button.click(
+        prepare_download,
+        inputs=[audio_input, original_text, corrected_text],
+        outputs=download_output
+    )
 
 demo.launch()