Create app.py

app.py

@@ -0,0 +1,42 @@
+# app.py
+import os
+import torch
+from transformers import pipeline
+import gradio as gr
+
+MODEL_ID = "EYEDOL/Yoruba-ASRNEW"
+
+# If you set HF_TOKEN as a secret in the Space (for private models), transformers auto-uses it.
+# Create pipeline. Use GPU if available in the Space.
+device = 0 if torch.cuda.is_available() else -1
+asr = pipeline("automatic-speech-recognition", model=MODEL_ID, device=device)
+
+def transcribe_from_file(audio_path):
+    """
+    audio_path: local filepath to recorded/uploaded audio (gradio provides wav/m4a etc.)
+    """
+    if not audio_path:
+        return "No audio provided."
+    # pipeline accepts filepath, numpy array, or list.
+    res = asr(audio_path)
+    return res.get("text", "")
+
+with gr.Blocks(title="Yoruba ASR Demo") as demo:
+    gr.Markdown("## Yoruba ASR — try microphone or upload an audio file 🎙️")
+    with gr.Tabs():
+        with gr.TabItem("Microphone"):
+            mic = gr.Audio(source="microphone", type="filepath", label="Record from mic")
+            mic_btn = gr.Button("Transcribe")
+            mic_out = gr.Textbox(label="Transcription")
+            mic_btn.click(fn=transcribe_from_file, inputs=mic, outputs=mic_out)
+        with gr.TabItem("Upload audio file"):
+            upload = gr.Audio(source="upload", type="filepath", label="Upload audio file")
+            up_btn = gr.Button("Transcribe file")
+            up_out = gr.Textbox(label="Transcription")
+            up_btn.click(fn=transcribe_from_file, inputs=upload, outputs=up_out)
+
+    gr.Markdown("**Notes:** If the model is private, set a `HF_TOKEN` secret in the Space settings. "
+                "For better speed, pick a GPU runtime (if available).")
+
+if __name__ == "__main__":
+    demo.launch()