feat: implement custom segmentation for Japanese language and update default source language

assets/sidecar/whisperx_worker.py

@@ -200,6 +200,10 @@ def build_segmentation_options(
     return options
 
 
+def should_apply_custom_segmentation(language: Optional[str]) -> bool:
+    return language == "ja"
+
+
 def load_wav_pcm_s16le(path: str) -> np.ndarray:
     with wave.open(path, "rb") as wav_file:
         channels = wav_file.getnchannels()
@@ -971,9 +975,6 @@ class WhisperXWorker:
             self.clear_device_resources(device="cuda")
 
         detected_language = str(result.get("language") or language or "unknown")
-        segmentation_options = build_segmentation_options(
-            detected_language, segmentation_overrides
-        )
         normalized_segments = normalize_segments(result.get("segments"))
 
         if not no_align and normalized_segments:
@@ -998,14 +999,17 @@ class WhisperXWorker:
                 del align_model
                 self.clear_device_resources(device="cuda")
             detected_language = str(aligned.get("language") or detected_language)
-            segmentation_options = build_segmentation_options(
-                detected_language, segmentation_overrides
-            )
-            normalized_segments = normalize_transcript_segments(
-                aligned.get("segments"),
-                detected_language,
-                segmentation_options,
-            )
+            if should_apply_custom_segmentation(detected_language):
+                segmentation_options = build_segmentation_options(
+                    detected_language, segmentation_overrides
+                )
+                normalized_segments = normalize_transcript_segments(
+                    aligned.get("segments"),
+                    detected_language,
+                    segmentation_options,
+                )
+            else:
+                normalized_segments = normalize_segments(aligned.get("segments"))
 
         emit_status(request_id, "finalizing")
         emit_progress(request_id, 96)

lib/services/whisper/whisper_service.dart

@@ -254,24 +254,8 @@ class WhisperService {
           'min_split_chars': 6,
           'prefer_punctuation_split': true,
         };
-      case 'zh':
-        return const <String, dynamic>{
-          'split_on_pause': true,
-          'pause_threshold_sec': 0.6,
-          'max_segment_duration_sec': 4.2,
-          'max_segment_chars': 26,
-          'min_split_chars': 4,
-          'prefer_punctuation_split': true,
-        };
       default:
-        return const <String, dynamic>{
-          'split_on_pause': true,
-          'pause_threshold_sec': 0.8,
-          'max_segment_duration_sec': 6.0,
-          'max_segment_chars': 42,
-          'min_split_chars': 10,
-          'prefer_punctuation_split': true,
-        };
+        return const <String, dynamic>{};
     }
   }
 

lib/ui/home_page.dart

@@ -47,7 +47,7 @@ class HomePage extends StatefulWidget {
 
 class _HomePageState extends State<HomePage> {
   String _selectedModel = AppConstants.defaultWhisperModel;
-  String _sourceVideoLanguage = 'en';
+  String _sourceVideoLanguage = 'ja';
   String _targetLanguage = 'zh';
   String _apiKey = '';
   String _targetModel = 'gemini-2.5-flash-lite';
@@ -147,7 +147,7 @@ class _HomePageState extends State<HomePage> {
           providerId: _llmProvider,
           apiKey: _apiKey,
           baseUrl: _llmBaseUrl,
-          sourceLanguage: 'en', // dummy
+          sourceLanguage: 'ja', // dummy
           targetLanguage: 'zh', // dummy
         ),
       );