From 6aac9bcae916da2326ae74b3ab14a23f33fcf8e6 Mon Sep 17 00:00:00 2001
From: owen <owen@p300.chcoilx1.pop.starlinkisp.net>
Date: Mon, 7 Sep 2026 23:38:56 -0500
Subject: [PATCH] fix: trim long chatterbox voice clips to 30s on upload

---
 ttstoy/ttstoy.py | 58 ++++++++++++++++++++++++++++++++++++++++++------
 1 file changed, 51 insertions(+), 7 deletions(-)

diff --git a/ttstoy/ttstoy.py b/ttstoy/ttstoy.py
index 297ecfa..7b040c7 100644
--- a/ttstoy/ttstoy.py
+++ b/ttstoy/ttstoy.py
@@ -169,6 +169,54 @@ def resolve_voice_for_mode(mode: str, voice_name: str, chatterbox_voices: list =
     return voice_name
 
 
+def _normalize_chatterbox_reference(
+    audio_seg, min_ms: int = 6000, max_ms: int = 30000
+):
+    """
+    Normalize a voice reference clip to a usable length for Chatterbox cloning.
+
+    - Clips shorter than ``min_ms`` are looped until they reach the minimum
+      duration required for a usable voice reference (>5s).
+    - Clips longer than ``max_ms`` are trimmed to the server's
+      ``max_reference_duration_sec`` limit, preferring a silence boundary near
+      the cut so words aren't chopped mid-syllable.
+
+    Returns the processed AudioSegment.
+    """
+    if audio_seg is None:
+        return audio_seg
+
+    if len(audio_seg) < min_ms:
+        loops_needed = (min_ms // len(audio_seg)) + 1 if len(audio_seg) > 0 else 1
+        audio_seg = audio_seg * loops_needed
+        log.info(f"[Chatterbox] Looped short clip {loops_needed}x to {len(audio_seg)/1000:.1f}s")
+
+    if len(audio_seg) > max_ms:
+        trimmed = audio_seg[:max_ms]
+        try:
+            from pydub.silence import detect_silence
+
+            search_start = max(0, max_ms - 1500)
+            silences = detect_silence(
+                audio_seg[search_start:max_ms],
+                min_silence_len=300,
+                silence_thresh=-40,
+            )
+            if silences:
+                cutoff_local = silences[0][0]
+                if cutoff_local > 100:
+                    trimmed = audio_seg[: search_start + cutoff_local]
+        except Exception:
+            pass
+        log.info(
+            f"[Chatterbox] Trimmed long clip from {len(audio_seg)/1000:.1f}s "
+            f"to {len(trimmed)/1000:.1f}s"
+        )
+        audio_seg = trimmed
+
+    return audio_seg
+
+
 def _split_voice_segments(text: str):
     """
     Split text on [mode|voice] tags into segments.
@@ -1591,17 +1639,13 @@ class TtsToy(Cog):
                     except Exception as e:
                         return await ctx.send(f"❌ Failed to download audio from URL: {e}")
 
-                # Ensure minimum duration (Chatterbox requires >5 seconds)
-                # Loop short clips until they're long enough
+                # Ensure reference clip is a usable length for voice cloning:
+                # loop short clips up to the minimum duration, trim long clips to the max
                 try:
                     from pydub import AudioSegment
                     import io as _io
                     audio_seg = AudioSegment.from_file(_io.BytesIO(audio_bytes))
-                    min_ms = 6000  # 6 seconds to be safe
-                    if len(audio_seg) < min_ms:
-                        loops_needed = (min_ms // len(audio_seg)) + 1
-                        audio_seg = audio_seg * loops_needed
-                        log.info(f"[Chatterbox] Looped short clip {loops_needed}x to {len(audio_seg)/1000:.1f}s")
+                    audio_seg = _normalize_chatterbox_reference(audio_seg)
                     # Always export as wav for best compatibility
                     buf = _io.BytesIO()
                     audio_seg.export(buf, format="wav")