v1.14.1: Add silence gap between TTS segments & fix Qwen3 TTS stitching

- CosyVoice: replace overlapping crossfade with fade-out + 100ms silence
  gap + fade-in for easier downstream splitting
- Qwen3 TTS: fix bare torch.cat in multi-emotion segment stitching,
  apply same fade-out + silence + fade-in approach

aiia_cosyvoice_nodes.py

@@ -586,41 +586,48 @@ class AIIA_CosyVoice_TTS:
         return segments
 
     @staticmethod
-    def _crossfade_segments(waveforms, sample_rate, crossfade_ms=50):
+    def _crossfade_segments(waveforms, sample_rate, crossfade_ms=50, silence_ms=100):
         """
-        Join waveform segments with a cosine crossfade to eliminate clicks/pops.
-        Each waveform is (1, N) or (C, N). crossfade_ms is the overlap duration.
-        Uses the same cosine fade approach as the Voice Conversion stitcher.
+        Join waveform segments with cosine fade-out / silence gap / fade-in.
+        Each waveform is (1, N) or (C, N).
+        - crossfade_ms: cosine fade duration applied to each segment's edge.
+        - silence_ms:   silence gap inserted between segments for easy splitting.
         """
         if not waveforms or len(waveforms) == 1:
             return torch.cat(waveforms, dim=-1) if waveforms else torch.zeros(1, 0)
 
         xfade_samples = int(sample_rate * crossfade_ms / 1000)
-
-        # Pre-compute fade curves (cosine)
-        t = torch.linspace(0, np.pi, xfade_samples, device=waveforms[0].device)
-        fade_out = 0.5 * (1.0 + torch.cos(t))  # 1 -> 0
-        fade_in  = 1.0 - fade_out               # 0 -> 1
+        silence_samples = int(sample_rate * silence_ms / 1000)
 
         result = waveforms[0]
         for i in range(1, len(waveforms)):
             curr = waveforms[i]
-            actual_xfade = min(xfade_samples, result.shape[-1], curr.shape[-1])
-            if actual_xfade < 2:
-                # Too short for crossfade, just concatenate
-                result = torch.cat([result, curr], dim=-1)
-                continue
 
-            # Trim fade curves to actual overlap
-            fo = fade_out[:actual_xfade]
-            fi = fade_in[:actual_xfade]
+            # --- Fade out the tail of previous segment ---
+            fo_len = min(xfade_samples, result.shape[-1])
+            if fo_len >= 2:
+                t_fo = torch.linspace(0, np.pi, fo_len, device=result.device)
+                fade_out = 0.5 * (1.0 + torch.cos(t_fo))  # 1 -> 0
+                result = torch.cat([
+                    result[..., :-fo_len],
+                    result[..., -fo_len:] * fade_out,
+                ], dim=-1)
 
-            overlap = result[..., -actual_xfade:] * fo + curr[..., :actual_xfade] * fi
-            result = torch.cat([
-                result[..., :-actual_xfade],
-                overlap,
-                curr[..., actual_xfade:],
-            ], dim=-1)
+            # --- Insert silence gap ---
+            channels = result.shape[0] if result.dim() >= 2 else 1
+            silence = torch.zeros(channels, silence_samples, device=result.device)
+
+            # --- Fade in the head of next segment ---
+            fi_len = min(xfade_samples, curr.shape[-1])
+            if fi_len >= 2:
+                t_fi = torch.linspace(0, np.pi, fi_len, device=curr.device)
+                fade_in = 1.0 - 0.5 * (1.0 + torch.cos(t_fi))  # 0 -> 1
+                curr = torch.cat([
+                    curr[..., :fi_len] * fade_in,
+                    curr[..., fi_len:],
+                ], dim=-1)
+
+            result = torch.cat([result, silence, curr], dim=-1)
 
         return result
 

aiia_qwen_nodes.py

@@ -401,8 +401,26 @@ class AIIA_Qwen_TTS:
                 sr_final = seg_result[0]["sample_rate"]
                 all_segment_wavs.append(seg_wav.squeeze(0))  # [C, T]
 
-            # Concatenate all segments
-            combined = torch.cat(all_segment_wavs, dim=-1)  # [C, T_total]
+            # Stitch segments: fade-out + silence gap + fade-in (same as CosyVoice)
+            XFADE_MS, SILENCE_MS = 50, 100
+            xfade_samples = int(sr_final * XFADE_MS / 1000)
+            silence_samples = int(sr_final * SILENCE_MS / 1000)
+            combined = all_segment_wavs[0]
+            for _si in range(1, len(all_segment_wavs)):
+                _curr = all_segment_wavs[_si]
+                # Fade out tail
+                fo_len = min(xfade_samples, combined.shape[-1])
+                if fo_len >= 2:
+                    t_fo = torch.linspace(0, np.pi, fo_len, device=combined.device)
+                    combined = torch.cat([combined[..., :-fo_len], combined[..., -fo_len:] * (0.5 * (1.0 + torch.cos(t_fo)))], dim=-1)
+                # Silence gap
+                silence = torch.zeros(combined.shape[0], silence_samples, device=combined.device)
+                # Fade in head
+                fi_len = min(xfade_samples, _curr.shape[-1])
+                if fi_len >= 2:
+                    t_fi = torch.linspace(0, np.pi, fi_len, device=_curr.device)
+                    _curr = torch.cat([_curr[..., :fi_len] * (1.0 - 0.5 * (1.0 + torch.cos(t_fi))), _curr[..., fi_len:]], dim=-1)
+                combined = torch.cat([combined, silence, _curr], dim=-1)
             return ({"waveform": combined.unsqueeze(0), "sample_rate": sr_final},)
 
         # --- Single segment path (original logic) ---

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "aiia"
 description = "Advanced AI Audio/Video toolkit for ComfyUI. Features Multi-Role Podcast/Dialogue Generation, High-Fidelity Voice Cloning (CosyVoice/VibeVoice), TTS, Media Management, and efficient Video tools."
-version = "1.14.0"
+version = "1.14.1"
 license = {file = "LICENSE"}
 readme = "README.md"
 authors = [