From ee96c05d154802916ec1163f1e17d11415f0d071 Mon Sep 17 00:00:00 2001
From: Jacob Schmidt <innovativestudios@outlook.com>
Date: Sun, 1 Mar 2026 00:39:11 -0600
Subject: [PATCH] Filter silent audio and blank placeholder transcripts in S2T

---
 Journal.Sidecar/LocalWhisperS2TService.cs | 37 +++++++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/Journal.Sidecar/LocalWhisperS2TService.cs b/Journal.Sidecar/LocalWhisperS2TService.cs
index 2f6a2f2..26fe532 100644
--- a/Journal.Sidecar/LocalWhisperS2TService.cs
+++ b/Journal.Sidecar/LocalWhisperS2TService.cs
@@ -14,6 +14,7 @@ public sealed class LocalWhisperS2TService : IS2TService, IDisposable
     private const int Channels = 1;
     private const int ChunkMs = 2000;
     private const int MaxBufferedItems = 256;
+    private const int SilenceRmsThreshold = 150;
 
     private readonly object _sync = new();
     private readonly object _segmentLock = new();
@@ -202,6 +203,9 @@ public sealed class LocalWhisperS2TService : IS2TService, IDisposable
             {
                 try
                 {
+                    if (IsLikelySilence(pcmChunk))
+                        continue;
+
                     using var pcmStream = new MemoryStream(pcmChunk, writable: false);
                     using var raw = new RawSourceWaveStream(pcmStream, waveFormat);
                     using var wavStream = new MemoryStream();
@@ -213,6 +217,8 @@ public sealed class LocalWhisperS2TService : IS2TService, IDisposable
                         var text = result.Text?.Trim();
                         if (string.IsNullOrWhiteSpace(text))
                             continue;
+                        if (IsPlaceholderTranscript(text))
+                            continue;
                         EnqueueTranscript(text);
                     }
                 }
@@ -253,6 +259,37 @@ public sealed class LocalWhisperS2TService : IS2TService, IDisposable
         return modelPath;
     }
 
+    private static bool IsLikelySilence(byte[] pcmChunk)
+    {
+        if (pcmChunk.Length < 2)
+            return true;
+
+        long sumSquares = 0;
+        int samples = pcmChunk.Length / 2;
+        for (int i = 0; i + 1 < pcmChunk.Length; i += 2)
+        {
+            short sample = (short)(pcmChunk[i] | (pcmChunk[i + 1] << 8));
+            sumSquares += (long)sample * sample;
+        }
+
+        if (samples <= 0)
+            return true;
+
+        var rms = Math.Sqrt(sumSquares / (double)samples);
+        return rms < SilenceRmsThreshold;
+    }
+
+    private static bool IsPlaceholderTranscript(string text)
+    {
+        var normalized = text.Trim();
+        if (!(normalized.StartsWith('[') && normalized.EndsWith(']')))
+            return false;
+
+        return normalized.Equals("[BLANK_AUDIO]", StringComparison.OrdinalIgnoreCase)
+            || normalized.Equals("[NO AUDIO]", StringComparison.OrdinalIgnoreCase)
+            || normalized.Equals("[SILENCE]", StringComparison.OrdinalIgnoreCase);
+    }
+
     private void EnqueueTranscript(string text)
     {
         _transcripts.Enqueue(text);