Add tabs for input, audio input

app.py

@@ -537,7 +537,7 @@ def hash_file(filepath):
         return hashlib.md5(f.read(1024*1024)).hexdigest()
 
 @spaces.GPU(duration=120)
-def process_pdf(pdf_file, url_input, yt_url_input, yt_cookies_file, translit_lang, translit_format, target_lang, max_text_char, repetition_penalty_val, last_source_hash, last_korean_words, progress=gr.Progress()):
+def process_pdf(pdf_file, url_input, audio_file_input, yt_url_input, yt_cookies_file, translit_lang, translit_format, target_lang, max_text_char, repetition_penalty_val, last_source_hash, last_korean_words, active_tab, progress=gr.Progress()):
     global tts, voice_style
     
     # Clean language choices from "Family - Language" to just "Language"
@@ -548,14 +548,20 @@ def process_pdf(pdf_file, url_input, yt_url_input, yt_cookies_file, translit_lan
         
     os.makedirs("log", exist_ok=True)
     
-    is_url = bool(url_input and url_input.strip())
-    is_youtube = bool(yt_url_input and yt_url_input.strip() and is_youtube_url(yt_url_input.strip()))
-    if pdf_file is None and not is_url and not is_youtube:
-        yield "<p>Please upload a PDF, enter a URL, or provide a YouTube link.</p>", None, None, "", "", [], None
+    # Determine input source based on active tab
+    is_url = (active_tab == "Website URL") and bool(url_input and url_input.strip())
+    is_youtube = (active_tab == "YouTube Link") and bool(yt_url_input and yt_url_input.strip() and is_youtube_url(yt_url_input.strip()))
+    is_audio_upload = (active_tab == "Upload Audio") and (audio_file_input is not None)
+    is_pdf = (active_tab == "Upload PDF") and (pdf_file is not None)
+    
+    if not is_url and not is_youtube and not is_audio_upload and not is_pdf:
+        yield "<p>Please provide input in the active tab.</p>", None, None, "", "", [], None
         return
         
     if is_youtube:
         current_source_hash = hashlib.md5(yt_url_input.strip().encode()).hexdigest()
+    elif is_audio_upload:
+        current_source_hash = hash_file(audio_file_input)
     elif is_url:
         current_source_hash = hashlib.md5(url_input.strip().encode()).hexdigest()
     else:
@@ -564,16 +570,6 @@ def process_pdf(pdf_file, url_input, yt_url_input, yt_cookies_file, translit_lan
     vocab_list = []
     extracted_audio_path = None
 
-    # if last_source_hash == current_source_hash and last_korean_words:
-    #     # Just run text-to-text LLM
-    #     progress(0.2, desc="Translating previously extracted vocabulary...")
-    #     korean_words = [item.get("korean") for item in last_korean_words if item.get("korean")]
-    #     for attempt in range(1, 4):
-    #         vocab_list = translate_vocabulary(korean_words, translit_lang, translit_format, target_lang, repetition_penalty_val)
-    #         if vocab_list:
-    #             break
-    # else:
-    
     try:
         if is_youtube:
             progress(0, desc="Downloading YouTube audio (first 5 min)...")
@@ -592,6 +588,18 @@ def process_pdf(pdf_file, url_input, yt_url_input, yt_cookies_file, translit_lan
             # Log the transcription
             with open("log/debug_yt_transcription.txt", "w", encoding="utf-8") as f:
                 f.write(f"Title: {yt_title}\n\n{content_text}")
+        elif is_audio_upload:
+            progress(0, desc="Transcribing uploaded audio with Cohere ASR...")
+            content_text = transcribe_audio_with_asr(audio_file_input)
+            images = []
+            extracted_audio_path = audio_file_input
+                
+            if not content_text.strip():
+                yield "<p>Could not transcribe any text from the uploaded audio.</p>", current_source_hash, None, "", "", [], extracted_audio_path
+                return
+                
+            with open("log/debug_audio_transcription.txt", "w", encoding="utf-8") as f:
+                f.write(content_text)
         elif is_url:
             progress(0, desc="Fetching Website...")
             content_text, images = extract_website_content(url_input.strip())
@@ -980,6 +988,16 @@ def get_example_pdf():
             print(f"Failed to download example PDF: {e}")
     return file_path if os.path.exists(file_path) else None
 
+def get_example_audio():
+    url = "https://raw.githubusercontent.com/ShayekhBinIslam/file-host/main/new_1min.wav"
+    file_path = "new_1min.wav"
+    if not os.path.exists(file_path):
+        try:
+            urllib.request.urlretrieve(url, file_path)
+        except Exception as e:
+            print(f"Failed to download example audio: {e}")
+    return file_path if os.path.exists(file_path) else None
+
 @spaces.GPU(duration=120)
 def process_pdf_force(partial_text, pdf_file, url_input, translit_lang, translit_format, target_lang, max_text_char, repetition_penalty_val, last_source_state, last_korean_words_state):
     """Force JSON generation using the current partial stream_box text."""
@@ -1062,6 +1080,7 @@ def process_pdf_force(partial_text, pdf_file, url_input, translit_lang, translit
 
 def create_demo():
     example_pdf = get_example_pdf()
+    example_audio = get_example_audio()
     
     custom_theme = gr.themes.Soft(
         primary_hue="violet",
@@ -1168,27 +1187,39 @@ def create_demo():
     
     with gr.Blocks(title="LocalDuo", theme=custom_theme, css=css) as demo:
         gr.Markdown("# 🇰🇷✨ LocalDuo - Learn Korean from PDFs, Websites & YouTube")
-        gr.Markdown("Enter a website URL 🌐, a YouTube link 🎬, or upload a Korean book PDF 📄. The app uses a **Vision-Language Model (VLM)** 🧠 to extract vocabulary from text and images, **ASR** 🎤 to transcribe YouTube audio, and a **Text-to-Speech (TTS)** engine 🗣️ to generate pronunciation audio.")
+        gr.Markdown("Enter a website URL 🌐, upload a PDF 📄, upload an audio file 🎵, or paste a YouTube link 🎬. The app uses a **Vision-Language Model (VLM)** 🧠, **ASR** 🎤, and **TTS** 🗣️ to generate vocabulary flashcards.")
+        
+        active_tab = gr.State("Website URL")
         
         with gr.Row():
             with gr.Column(scale=1):
-                # url_input = gr.Textbox(label="Enter a Website URL 🌐", placeholder=r"e.g. https://storykorean.com/stories?level=beginner&story=tiger", value=r"https://storykorean.com/stories?level=beginner&story=tiger")
-                # https://www.bbc.com/korean/articles/c5yz89k5dw0o
-                # https://www.bbc.com/korean/articles/cn0p7rkvxdgo
-                # https://www.koreanstudyjunkie.com/post/korean-reading-exercise-for-all-levels-beginner-intermediate-advanced
-                url_input = gr.Textbox(label="Enter a Website URL 🌐", 
-                placeholder=r"e.g. # https://www.bbc.com/korean/articles/cn0p7rkvxdgo", 
-                value=r"https://www.bbc.com/korean/articles/cn0p7rkvxdgo")
+                with gr.Tabs() as input_tabs:
+                    with gr.Tab("Website URL", id="tab_url") as tab_url:
+                        url_input = gr.Textbox(label="Enter a Website URL 🌐", 
+                        placeholder=r"e.g. https://www.bbc.com/korean/articles/cn0p7rkvxdgo", 
+                        value=r"https://www.bbc.com/korean/articles/cn0p7rkvxdgo")
+                    
+                    with gr.Tab("Upload PDF", id="tab_pdf") as tab_pdf:
+                        pdf_input = gr.File(label="Upload Book PDF 📚", file_types=[".pdf"], value=example_pdf)
+                    
+                    with gr.Tab("Upload Audio", id="tab_audio") as tab_audio:
+                        audio_file_input = gr.File(label="Upload Audio File 🎵", file_types=[".wav", ".mp3", ".m4a", ".ogg", ".flac", ".opus", ".webm"], value=example_audio)
+                        gr.Markdown("*Upload a Korean audio file. It will be transcribed using Cohere ASR and vocabulary will be extracted from the transcript.*", elem_classes=["hint-text"])
+                    
+                    with gr.Tab("YouTube Link", id="tab_yt") as tab_yt:
+                        yt_url_input = gr.Textbox(label="Enter a YouTube Link 🎬", 
+                        placeholder=r"e.g. https://www.youtube.com/watch?v=...", 
+                        value="https://www.youtube.com/watch?v=9Nj7l73PBWE",
+                        info="Audio from the first 5 minutes will be transcribed using Cohere ASR")
+                        yt_cookies_input = gr.File(label="YouTube Cookies (cookies.txt)", file_types=[".txt"], value=None,
+                            type="filepath")
+                        gr.Markdown("*Optional. Helps bypass YouTube bot detection. Install the [cookies.txt](https://addons.mozilla.org/firefox/addon/cookies-txt/) extension, go to youtube.com while logged in, click the extension → 'Current Site' to export.*", elem_classes=["hint-text"])
                 
-                yt_url_input = gr.Textbox(label="Or Enter a YouTube Link 🎬", 
-                placeholder=r"e.g. https://www.youtube.com/watch?v=...", 
-                value="https://www.youtube.com/watch?v=9Nj7l73PBWE",
-                info="Audio from the first 5 minutes will be transcribed using Cohere ASR")
-                yt_cookies_input = gr.File(label="YouTube Cookies (cookies.txt)", file_types=[".txt"], value=None,
-                    type="filepath")
-                gr.Markdown("*Optional. Helps bypass YouTube bot detection. Install the [cookies.txt](https://addons.mozilla.org/firefox/addon/cookies-txt/) extension, go to youtube.com while logged in, click the extension → 'Current Site' to export.*", elem_classes=["hint-text"])
-
-                pdf_input = gr.File(label="Or Upload Book PDF 📚", file_types=[".pdf"], value=example_pdf)
+                # Track active tab
+                tab_url.select(fn=lambda: "Website URL", inputs=None, outputs=active_tab)
+                tab_pdf.select(fn=lambda: "Upload PDF", inputs=None, outputs=active_tab)
+                tab_audio.select(fn=lambda: "Upload Audio", inputs=None, outputs=active_tab)
+                tab_yt.select(fn=lambda: "YouTube Link", inputs=None, outputs=active_tab)
                 
                 gr.Markdown("### ⚙️ Customization Settings")
                 translit_lang = gr.Dropdown(
@@ -1217,7 +1248,7 @@ def create_demo():
                 with gr.Accordion("📄 Extracted Source Content", open=True):
                     extracted_text_box = gr.Textbox(label="Extracted Text", lines=10, max_lines=15, interactive=False)
                     extracted_images_gallery = gr.Gallery(label="Extracted Images", columns=4, height="auto", object_fit="contain")
-                    extracted_audio_player = gr.Audio(label="Extracted Audio (YouTube)", type="filepath", interactive=False)
+                    extracted_audio_player = gr.Audio(label="Extracted Audio (YouTube / Uploaded)", type="filepath", interactive=False)
                 
         last_source_state = gr.State(None)
         last_korean_words_state = gr.State(None)
@@ -1229,7 +1260,7 @@ def create_demo():
 
         generate_event = submit_btn.click(
             fn=process_pdf, 
-            inputs=[pdf_input, url_input, yt_url_input, yt_cookies_input, translit_lang, translit_format, target_lang, max_text_char_input, repetition_penalty_input, last_source_state, last_korean_words_state], 
+            inputs=[pdf_input, url_input, audio_file_input, yt_url_input, yt_cookies_input, translit_lang, translit_format, target_lang, max_text_char_input, repetition_penalty_input, last_source_state, last_korean_words_state, active_tab], 
             outputs=[output_html, last_source_state, last_korean_words_state, stream_box, extracted_text_box, extracted_images_gallery, extracted_audio_player]
         )