Create app.py

app.py

@@ -0,0 +1,56 @@
+import os
+import yt_dlp
+import gradio as gr
+from transformers import pipeline
+
+# Load your HF token safely (set this in your Hugging Face Space secrets as HF_TOKEN)
+HF_TOKEN = os.environ.get("HF_TOKEN", "your_token_here")
+
+# Choose a small Burmese-capable ASR model
+MODEL_NAME = "chuuhtetnaing/whisper-tiny-myanmar"
+
+# Initialize pipeline
+asr_pipeline = pipeline(
+    task="automatic-speech-recognition",
+    model=MODEL_NAME,
+    use_auth_token=HF_TOKEN
+)
+
+def download_audio(youtube_url, out_path="audio.wav"):
+    """Download and convert YouTube video to audio wav"""
+    ydl_opts = {
+        "format": "bestaudio/best",
+        "outtmpl": "audio.%(ext)s",
+        "postprocessors": [{
+            "key": "FFmpegExtractAudio",
+            "preferredcodec": "wav",
+            "preferredquality": "192",
+        }],
+    }
+    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
+        ydl.download([youtube_url])
+
+    # yt-dlp usually saves as audio.wav
+    return "audio.wav"
+
+def transcribe(youtube_url):
+    """Main function: download + transcribe"""
+    try:
+        audio_file = download_audio(youtube_url)
+        result = asr_pipeline(audio_file, chunk_length_s=30)  # handles long audio in 30s chunks
+        text = result["text"]
+        return text
+    except Exception as e:
+        return f"❌ Error: {str(e)}"
+
+# Gradio interface
+demo = gr.Interface(
+    fn=transcribe,
+    inputs=gr.Textbox(label="YouTube URL", placeholder="Paste your YouTube link here..."),
+    outputs=gr.Textbox(label="Transcribed Text (Burmese)"),
+    title="🎙️ Burmese Speech-to-Text (YouTube)",
+    description="Paste a YouTube link with Burmese audio (up to ~20 minutes). It will transcribe the speech into text."
+)
+
+if __name__ == "__main__":
+    demo.launch()