Update src/gradio_app.py

src/gradio_app.py

@@ -6,6 +6,7 @@ import tempfile
 import re as regex
 import yt_dlp
 import glob
+import random
 
 # --- Configuration ---
 MAX_SIZE_MB = "50"
@@ -13,6 +14,11 @@ MAX_SECONDS = 60
 LIBRE_API_KEY = "xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx"
 TRANSLATE_URL = "https://imsidag-community-libretranslate-kabyle.hf.space/translate"
 
+# --- Dataset Configuration ---
+DATASET_REPO = "boffire/kabyle-synth-voice"
+DATASET_AUDIO_BASE_URL = f"https://huggingface.co/datasets/{DATASET_REPO}/resolve/main/audio"
+DATASET_API_TREE_URL = f"https://huggingface.co/api/datasets/{DATASET_REPO}/tree/main/audio"
+
 # --- Translation Logic ---
 def translate_to_english(text):
     if not text or any(symbol in text for symbol in ["⚠️", "❌"]):
@@ -81,6 +87,52 @@ def download_soundcloud_audio(url: str) -> str:
     except Exception as e:
         raise RuntimeError(f"yt-dlp failed: {str(e)}")
 
+# --- Dataset Random Sample Logic ---
+_audio_files_cache = None
+
+def get_dataset_audio_files():
+    """Fetch the list of audio files from the dataset API (cached)."""
+    global _audio_files_cache
+    if _audio_files_cache is not None:
+        return _audio_files_cache
+
+    try:
+        resp = requests.get(DATASET_API_TREE_URL, timeout=15)
+        resp.raise_for_status()
+        items = resp.json()
+        # Filter only .wav files and extract filenames
+        files = [
+            item["path"].replace("audio/", "")
+            for item in items
+            if item.get("type") == "file" and item["path"].endswith(".wav")
+        ]
+        _audio_files_cache = files
+        return files
+    except Exception as e:
+        raise RuntimeError(f"Failed to fetch dataset file list: {e}")
+
+def download_random_dataset_sample() -> str:
+    """Pick a random audio file from the dataset and download it."""
+    files = get_dataset_audio_files()
+    if not files:
+        raise RuntimeError("No audio files found in the dataset.")
+
+    filename = random.choice(files)
+    file_url = f"{DATASET_AUDIO_BASE_URL}/{filename}"
+    tmp_dir = tempfile.gettempdir()
+    local_path = os.path.join(tmp_dir, f"dataset_{filename}")
+
+    # Download the file
+    try:
+        resp = requests.get(file_url, timeout=30, stream=True)
+        resp.raise_for_status()
+        with open(local_path, "wb") as f:
+            for chunk in resp.iter_content(chunk_size=8192):
+                f.write(chunk)
+        return local_path
+    except Exception as e:
+        raise RuntimeError(f"Failed to download {filename}: {e}")
+
 # --- Unified Processing Logic ---
 def process_audio(audio_file):
     """Handles validation -> Transcription -> Translation."""
@@ -124,6 +176,24 @@ def process_soundcloud(url):
 
     return transcript, translation
 
+def process_random_dataset():
+    """Downloads a random sample from the dataset and runs ASR."""
+    try:
+        audio_path = download_random_dataset_sample()
+    except Exception as e:
+        return f"❌ Dataset Error: {str(e)}", ""
+
+    transcript, translation = process_audio(audio_path)
+
+    # Cleanup temp file
+    try:
+        if os.path.exists(audio_path):
+            os.remove(audio_path)
+    except Exception:
+        pass
+
+    return transcript, translation
+
 # --- Build Gradio UI ---
 with gr.Blocks(title="🎙️ Mmeslay") as demo:
     gr.Markdown(
@@ -132,7 +202,7 @@ with gr.Blocks(title="🎙️ Mmeslay") as demo:
         ### Kabyle ASR & Translation
         *Powered by Squeezeformer (ASR) and LibreTranslate (NMT)*
 
-        Upload a Kabyle audio file, record directly, **or paste a SoundCloud link** to get a transcript and English translation.
+        Upload a Kabyle audio file, record directly, **or pick a random sample** from the Kabyle Synth Voice dataset to get a transcript and English translation.
         """
     )
 
@@ -170,28 +240,50 @@ with gr.Blocks(title="🎙️ Mmeslay") as demo:
             inputs=audio_input,
         )
 
-    with gr.Tab("🎵 SoundCloud Link"):
+    with gr.Tab("🎲 Random Dataset Sample"):
         with gr.Row():
             with gr.Column(scale=1):
-                sc_url = gr.Textbox(
-                    label="SoundCloud URL",
-                    placeholder="https://soundcloud.com/artist/track",
-                    lines=1
+                gr.Markdown(
+                    """
+                    Click the button below to fetch a **random audio sample** from the [Kabyle Synth Voice](https://huggingface.co/datasets/boffire/kabyle-synth-voice) dataset.
+                    """
                 )
-                sc_btn = gr.Button("🚀 Download & Transcribe", variant="primary", size="lg")
+                random_btn = gr.Button("🎲 Pick Random & Transcribe", variant="primary", size="lg")
+                dataset_status = gr.Textbox(label="Status", interactive=False, value="Ready")
 
             with gr.Column(scale=2):
-                text_output_2 = gr.Textbox(label="Transcription (Kabyle)", lines=5)
-                translation_output_2 = gr.Textbox(
+                text_output_3 = gr.Textbox(label="Transcription (Kabyle)", lines=5)
+                translation_output_3 = gr.Textbox(
                     label="LibreTranslate (English)",
                     lines=5,
                     placeholder="English LibreTranslate translation will appear here..."
                 )
 
-        sc_btn.click(
-            fn=process_soundcloud,
-            inputs=sc_url,
-            outputs=[text_output_2, translation_output_2],
+        def process_random_with_status():
+            # Update status
+            yield "⏳ Fetching random sample...", "", ""
+            try:
+                audio_path = download_random_dataset_sample()
+            except Exception as e:
+                yield f"❌ Dataset Error: {str(e)}", "", ""
+                return
+
+            yield "⏳ Transcribing...", "", ""
+            transcript, translation = process_audio(audio_path)
+
+            # Cleanup temp file
+            try:
+                if os.path.exists(audio_path):
+                    os.remove(audio_path)
+            except Exception:
+                pass
+
+            yield "✅ Done!", transcript, translation
+
+        random_btn.click(
+            fn=process_random_with_status,
+            inputs=[],
+            outputs=[dataset_status, text_output_3, translation_output_3],
         )
 
     gr.Markdown(