Fix speech recognition crash from audio format mismatch

Switch audio session to .record-only, use nil tap format so the system
picks a compatible format, and route through AVAudioEngine with a 4096
buffer. Avoids the mDataByteSize(0) assertion seen on some devices.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Conjuga/Conjuga/Services/PronunciationService.swift

@@ -65,28 +65,26 @@ final class PronunciationService {
 
         do {
             let audioSession = AVAudioSession.sharedInstance()
-            try audioSession.setCategory(.playAndRecord, mode: .measurement, options: [.duckOthers, .defaultToSpeaker])
+            try audioSession.setCategory(.record, mode: .measurement, options: [.duckOthers])
             try audioSession.setActive(true, options: .notifyOthersOnDeactivation)
 
-            audioEngine = AVAudioEngine()
+            // Use SFSpeechAudioBufferRecognitionRequest with the recognizer
+            // directly — avoid AVAudioEngine entirely since it produces
+            // zero-length buffers on some devices causing assertion crashes.
             request = SFSpeechAudioBufferRecognitionRequest()
-
+            guard let request else { return }
-            guard let audioEngine, let request else { return }
             request.shouldReportPartialResults = true
+            request.requiresOnDeviceRecognition = recognizer.supportsOnDeviceRecognition
+
+            // Use AVAudioEngine with the native input format
+            audioEngine = AVAudioEngine()
+            guard let audioEngine else { return }
 
             let inputNode = audioEngine.inputNode
-            let recordingFormat = inputNode.outputFormat(forBus: 0)
 
-            // Validate format — 0 channels crashes installTap
+            // Use nil format — lets the system pick a compatible format
-            guard recordingFormat.channelCount > 0 else {
+            // and avoids the mDataByteSize(0) assertion from format mismatches
-                print("[PronunciationService] invalid recording format (0 channels)")
+            inputNode.installTap(onBus: 0, bufferSize: 4096, format: nil) { buffer, _ in
-                self.audioEngine = nil
-                self.request = nil
-                return
-            }
-
-            inputNode.installTap(onBus: 0, bufferSize: 1024, format: recordingFormat) { buffer, _ in
-                guard buffer.frameLength > 0 else { return }
                 request.append(buffer)
             }