perf: reduce prompt relay and audio ui overhead

load_audio_ui.py

@@ -1,8 +1,16 @@
 import folder_paths
+import numpy as np
 import os
 import torch
 import av
 
+DEFAULT_SAMPLE_RATE = 44100
+DEFAULT_SILENCE_SECONDS = 1
+
+
+def build_silence(sample_rate: int = DEFAULT_SAMPLE_RATE, seconds: int = DEFAULT_SILENCE_SECONDS, channels: int = 2) -> torch.Tensor:
+    return torch.zeros((channels, int(sample_rate * seconds)), dtype=torch.float32)
+
 def f32_pcm(wav: torch.Tensor) -> torch.Tensor:
     """Convert audio to float 32 bits PCM format."""
     if wav.dtype.is_floating_point:
@@ -23,18 +31,17 @@ def load_audio_file(filepath: str) -> tuple[torch.Tensor, int]:
         sr = stream.codec_context.sample_rate
         n_channels = stream.channels
 
-        frames = []
+        frame_arrays = []
         for frame in af.decode(streams=stream.index):
-            buf = torch.from_numpy(frame.to_ndarray())
-            if buf.shape[0] != n_channels:
-                buf = buf.view(-1, n_channels).t()
+            array = frame.to_ndarray()
+            if array.shape[0] != n_channels:
+                array = array.reshape(-1, n_channels).T
+            frame_arrays.append(array)
 
-            frames.append(buf)
-
-        if not frames:
+        if not frame_arrays:
             raise ValueError("No audio frames decoded.")
 
-        wav = torch.cat(frames, dim=1)
+        wav = torch.from_numpy(np.concatenate(frame_arrays, axis=1))
         wav = f32_pcm(wav)
         return wav, sr
 
@@ -99,17 +106,16 @@ class LoadAudioUI:
             missing_info = audio if audio != "none" else "None selected"
             print(f"!!! [LoadAudioUI] Warning: Audio file '{missing_info}' not found. Outputting 1 second of silence.")
 
-            sample_rate = 44100
-            # 1 second of silence (stereo) -> shape [channels, time]
-            waveform = torch.zeros((2, 44100))
+            sample_rate = DEFAULT_SAMPLE_RATE
+            waveform = build_silence(sample_rate)
         else:
             try:
                 waveform, sample_rate = load_audio_file(audio_path)
             except Exception as e:
                 # If decoding fails for any reason, fallback to silence rather than crashing the workflow
                 print(f"!!! [LoadAudioUI] Error decoding {audio}: {e}. Falling back to silence.")
-                sample_rate = 44100
-                waveform = torch.zeros((2, 44100))
+                sample_rate = DEFAULT_SAMPLE_RATE
+                waveform = build_silence(sample_rate)
 
         # Convert seconds to frames
         start_frame = int(start_time * sample_rate)

patches.py

@@ -1,9 +1,12 @@
 import logging
+import os
 import types
 
 import comfy.ldm.modules.attention
 
 log = logging.getLogger(__name__)
+_DEBUG_PROMPT_RELAY = os.environ.get("PROMPT_RELAY_DEBUG", "").strip().lower() in {"1", "true", "yes", "on"}
+_DEBUG_LOG_PATH = os.path.join(os.path.dirname(__file__), "debug_prompt_relay.log")
 
 
 def _masked_attention(q, k, v, heads, mask, transformer_options={}, **kwargs):
@@ -73,10 +76,10 @@ def _make_masked_override(prev_override):
 
 
 def debug_log(msg):
+    if not _DEBUG_PROMPT_RELAY:
+        return
     try:
-        import os
-        log_path = os.path.join(os.path.dirname(__file__), "debug_prompt_relay.log")
-        with open(log_path, "a", encoding="utf-8") as f:
+        with open(_DEBUG_LOG_PATH, "a", encoding="utf-8") as f:
             f.write(msg + "\n")
     except Exception:
         pass