Fix make_dub filter_complex: don't include silent base as filter, only delay TTS segments

app.py

@@ -47,7 +47,6 @@ def load_asr():
 
 def transcribe(wav):
     processor, model = load_asr()
-    # Use soundfile to load (avoids torchcodec issues)
     audio, sr = sf.read(str(wav))
     if len(audio.shape) > 1:
         audio = audio.mean(axis=1)
@@ -112,11 +111,9 @@ def load_tts():
 
 def tts(text, wav):
     tokenizer, model = load_tts()
-    # Add attention_mask to avoid warning
     inputs = tokenizer(text, return_tensors="pt", padding=True)
     with torch.inference_mode():
         audio = model(**inputs).waveform.cpu()
-    # Use soundfile instead of torchaudio.save (avoids torchcodec)
     sf.write(str(wav), audio.squeeze().numpy(), model.config.sampling_rate)
 
 def fit_audio(src, dst, target):
@@ -140,12 +137,14 @@ def make_dub(ms, total, work):
     weights = [max(1,len(re.sub(r'\s+','',x))) for x in parts]
     total_w = sum(weights)
 
+    # Create silent base
     silent = work/"silent.wav"
     run(["ffmpeg","-y","-f","lavfi","-i","anullsrc=r=24000:cl=mono",
          "-t",f"{total:.3f}","-ar","24000","-ac","1",str(silent)])
 
-    inputs = ["-i",str(silent)]
-    labels = ["[0:a]"]
+    # Generate and fit each TTS segment
+    inputs = ["-i", str(silent)]
+    filter_parts = []
     start = 0.0
 
     for i, part in enumerate(parts, 1):
@@ -153,13 +152,16 @@ def make_dub(ms, total, work):
         raw, fitted = work/f"tts_{i}.wav", work/f"fit_{i}.wav"
         tts(part, raw)
         fit_audio(raw, fitted, slot)
-        inputs += ["-i",str(fitted)]
+        inputs += ["-i", str(fitted)]
         delay = int(start*1000)
-        labels.append(f"[{i}:a]adelay={delay}|{delay}[d{i}]")
+        filter_parts.append(f"[{i}:a]adelay={delay}|{delay}[d{i}]")
         start += slot
 
-    mix = "[0:a]" + "".join(f"[d{i}]" for i in range(1,len(parts)+1))
-    filt = ";".join(labels) + f";{mix}amix=inputs={len(parts)+1}:duration=first[a]"
+    # Build filter_complex: delay each TTS segment, then mix with silent base
+    # Input 0 = silent, Input 1..n = TTS segments
+    mix_inputs = "[0:a]" + "".join(f"[d{i}]" for i in range(1, len(parts)+1))
+    filt = ";".join(filter_parts) + f";{mix_inputs}amix=inputs={len(parts)+1}:duration=first[a]"
+    
     out = work/"dub.wav"
     run(["ffmpeg","-y",*inputs,"-filter_complex",filt,"-map","[a]",
          "-ar","24000","-ac","1",str(out)])