increase streaming timeout to 120s; add plain-text + CSV export

app.py

@@ -1,5 +1,6 @@
 import io
 import os
+import tempfile
 import time
 import datetime
 import threading
@@ -20,7 +21,7 @@ os.environ.setdefault("HF_XET_HIGH_PERFORMANCE", "1")
 N_CPUS = os.cpu_count() or 2
 BATCH_SIZE = 64  # max clips per ONNX forward pass
 API_TIMEOUT = 12   # seconds per datasets-server attempt (one try per endpoint)
-STREAMING_TIMEOUT = 45  # seconds before giving up on streaming fallback
+STREAMING_TIMEOUT = 120  # seconds before giving up on streaming fallback
 
 
 def _ts() -> str:
@@ -220,7 +221,7 @@ def identify_speakers(
 ):
     repos = [r.strip() for r in repo_ids_text.strip().splitlines() if r.strip()]
     if not repos:
-        return pd.DataFrame(), "No repos provided.", ""
+        return pd.DataFrame(), "No repos provided.", "", "", None
 
     token = hf_token.strip() or os.environ.get("HF_TOKEN") or None
 
@@ -258,7 +259,7 @@ def identify_speakers(
     log.append(f"[{_ts()}] Phase 1 done: {ok} ok, {failed} failed, phase_total={int((time.time()-t_total)*1000)}ms")
 
     if not repo_arrays:
-        return pd.DataFrame(), "No audio downloaded.", "\n".join(log)
+        return pd.DataFrame(), "No audio downloaded.", "\n".join(log), "", None
 
     # ── Phase 2: batch embed all clips in one shot ───────────────────────────
     log.append(f"[{_ts()}] --- Phase 2: batch embed ---")
@@ -330,7 +331,20 @@ def identify_speakers(
     summary = f"✅ {len(repo_names)} books → {n_speakers} unique speakers  ({total_ms/1000:.1f}s total)"
     log.append(f"[{_ts()}] === DONE: {total_ms}ms total ===")
 
-    return df, summary, "\n".join(log)
+    # Plain-text copy-friendly output (space-separated, matches log format)
+    text_lines = []
+    for _, r in df.iterrows():
+        text_lines.append(
+            f"{r['dataset']}  {r['speaker_id']}  {r['books_with_speaker']}  {r['intra_sim']}  {r['closest_match']}"
+        )
+    plain_text = "\n".join(text_lines)
+
+    # CSV file for download
+    tmp = tempfile.NamedTemporaryFile(mode="w", suffix=".csv", delete=False, encoding="utf-8")
+    df.to_csv(tmp.name, index=False)
+    tmp.close()
+
+    return df, summary, "\n".join(log), plain_text, tmp.name
 
 
 DESCRIPTION = """
@@ -405,12 +419,20 @@ with gr.Blocks(title="Speaker Identifier") as demo:
         headers=["dataset", "speaker_id", "books_with_speaker", "intra_sim", "closest_match"],
         wrap=True,
     )
+    with gr.Row():
+        text_out = gr.Textbox(
+            label="Plain text (copy-friendly)",
+            interactive=False,
+            lines=12,
+            info="dataset  speaker_id  n_books  intra_sim  closest_match",
+        )
+        csv_out = gr.File(label="Download CSV", file_types=[".csv"])
     errors_out = gr.Textbox(label="Errors / Timing", interactive=False)
 
     run_btn.click(
         identify_speakers,
         inputs=[repo_input, samples, audio_sec, threshold, hf_token],
-        outputs=[table_out, summary_out, errors_out],
+        outputs=[table_out, summary_out, errors_out, text_out, csv_out],
     )
 
 demo.launch()