From ff4f90612856b4f4ebf7563ada3d20039cee73a1 Mon Sep 17 00:00:00 2001
From: Trey t <treytartt@fastmail.com>
Date: Mon, 13 Apr 2026 19:00:51 -0500
Subject: [PATCH] Fix crash from zero-length audio buffers in speech
 recognition
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Guard against empty audio buffers before appending to speech
recognition request — AVAudioBuffer asserts non-zero data size.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 Conjuga/Conjuga/Services/PronunciationService.swift | 1 +
 1 file changed, 1 insertion(+)

diff --git a/Conjuga/Conjuga/Services/PronunciationService.swift b/Conjuga/Conjuga/Services/PronunciationService.swift
index 0177bb9..fee2995 100644
--- a/Conjuga/Conjuga/Services/PronunciationService.swift
+++ b/Conjuga/Conjuga/Services/PronunciationService.swift
@@ -86,6 +86,7 @@ final class PronunciationService {
             }
 
             inputNode.installTap(onBus: 0, bufferSize: 1024, format: recordingFormat) { buffer, _ in
+                guard buffer.frameLength > 0 else { return }
                 request.append(buffer)
             }