AudioGen: Auto-Trim Stille für besseres Looping

Neue Option 'trim_silence' (Ja/Nein) im AudioGen Node.
Schneidet automatisch Stille am Anfang und Ende ab
(Threshold -40dB, 10ms Padding). Default: Ja.
Besseres Loop-Verhalten da keine Pause mehr am Anfang.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

comfyui-audio/comfyui_audiocraft/nodes.py

@@ -144,6 +144,22 @@ class MusicGenLongNode:
         return (out_path + ".wav",)
 
 
+def _trim_silence(wav, sample_rate, threshold_db=-40.0, pad_ms=10):
+    """Schneidet Stille am Anfang und Ende ab"""
+    threshold = 10 ** (threshold_db / 20.0)
+    # Mono oder Stereo: max über alle Kanäle
+    amplitude = wav.abs().max(dim=0).values
+    above = (amplitude > threshold).nonzero(as_tuple=True)[0]
+    if len(above) == 0:
+        return wav
+    start = max(0, above[0].item() - int(pad_ms * sample_rate / 1000))
+    end = min(wav.shape[1], above[-1].item() + int(pad_ms * sample_rate / 1000))
+    trimmed = wav[:, start:end]
+    if trimmed.shape[1] != wav.shape[1]:
+        print(f"[Trim] {wav.shape[1]/sample_rate:.2f}s → {trimmed.shape[1]/sample_rate:.2f}s (Stille entfernt)")
+    return trimmed
+
+
 def _next_number(output_dir, base_name):
     """Findet die nächste fortlaufende Nummer für einen Dateinamen"""
     pattern = os.path.join(output_dir, f"{base_name}_*.wav")
@@ -168,6 +184,7 @@ class AudioGenNode:
                 "temperature": ("FLOAT",  {"default": 1.0,  "min": 0.1,  "max": 2.0,  "step": 0.05}),
                 "cfg_coef":    ("FLOAT",  {"default": 3.0,  "min": 0.5,  "max": 10.0, "step": 0.5}),
                 "top_k":       ("INT",    {"default": 250,  "min": 1,    "max": 1000}),
+                "trim_silence": (["Ja", "Nein"], {"default": "Ja"}),
             }
         }
 
@@ -177,7 +194,7 @@ class AudioGenNode:
     CATEGORY = "AudioCraft"
     OUTPUT_NODE = True
 
-    def generate(self, prompt, name, duration, seed, temperature, cfg_coef, top_k):
+    def generate(self, prompt, name, duration, seed, temperature, cfg_coef, top_k, trim_silence):
         from audiocraft.models import AudioGen
         from audiocraft.data.audio import audio_write
 
@@ -195,6 +212,10 @@ class AudioGenNode:
 
         print(f"[AudioGen] Generating: {prompt}")
         wav = ag.generate([prompt])
+        result = wav[0].cpu()
+
+        if trim_silence == "Ja":
+            result = _trim_silence(result, ag.sample_rate)
 
         output_dir = "/app/ComfyUI/output/audio"
         os.makedirs(output_dir, exist_ok=True)
@@ -202,8 +223,8 @@ class AudioGenNode:
         num = _next_number(output_dir, base_name)
         filename = f"{base_name}_{num:03d}"
         out_path = os.path.join(output_dir, filename)
-        audio_write(out_path, wav[0].cpu(), ag.sample_rate, strategy="loudness")
-        print(f"[AudioGen] Saved: {filename}.wav")
+        audio_write(out_path, result, ag.sample_rate, strategy="loudness")
+        print(f"[AudioGen] Saved: {filename}.wav ({result.shape[1]/ag.sample_rate:.2f}s)")
 
         return (out_path + ".wav",)
 

comfyui-audio/workflow_sfx_optimized.json

@@ -23,7 +23,8 @@
         "fixed",
         0.8,
         4.5,
-        200
+        200,
+        "Ja"
       ]
     },
     {
@@ -79,7 +80,8 @@
         "fixed",
         1.0,
         3.5,
-        250
+        250,
+        "Ja"
       ]
     },
     {
@@ -135,7 +137,8 @@
         "fixed",
         1.0,
         3.0,
-        250
+        250,
+        "Ja"
       ]
     },
     {