diff --git a/app/Services/GoogleCloudService.php b/app/Services/GoogleCloudService.php
index 832a2e6e..cd5fc4d5 100644
--- a/app/Services/GoogleCloudService.php
+++ b/app/Services/GoogleCloudService.php
@@ -432,9 +432,10 @@ private function parseDiarizationResult(array $operationResult): ?array
         }
 
         // word-level 화자 정보를 세그먼트로 그룹핑
+        // Google STT의 SentencePiece 토크나이저: ▁(U+2581)는 새 단어 시작 표시
         $segments = [];
         $currentSpeaker = null;
-        $currentWords = [];
+        $currentTokens = [];
         $segmentStartMs = 0;
 
         foreach ($words as $word) {
@@ -443,36 +444,39 @@ private function parseDiarizationResult(array $operationResult): ?array
             $startMs = $this->parseGoogleTimeToMs($word['startTime'] ?? '0s');
             $endMs = $this->parseGoogleTimeToMs($word['endTime'] ?? '0s');
 
-            // 언더스코어 노이즈 제거 (단어 앞뒤/내부 모두)
-            $cleanWord = str_replace('_', '', $wordText);
-            if (trim($cleanWord) === '') {
+            // SentencePiece: ▁(U+2581) 또는 _로 시작하면 새 단어
+            $isNewWord = preg_match('/^[\x{2581}_]/u', $wordText);
+
+            // 모든 구분자 문자 제거: _(U+005F), ▁(U+2581)
+            $cleanToken = preg_replace('/[\x{2581}_]/u', '', $wordText);
+            if (trim($cleanToken) === '') {
                 continue;
             }
 
-            if ($speakerTag !== $currentSpeaker && $currentSpeaker !== null && ! empty($currentWords)) {
+            if ($speakerTag !== $currentSpeaker && $currentSpeaker !== null && ! empty($currentTokens)) {
                 $segments[] = [
                     'speaker_name' => '화자 ' . $currentSpeaker,
                     'speaker_label' => (string) $currentSpeaker,
-                    'text' => $this->cleanSttText(implode(' ', $currentWords)),
+                    'text' => $this->joinSentencePieceTokens($currentTokens),
                     'start_time_ms' => $segmentStartMs,
                     'end_time_ms' => $startMs,
                     'is_manual_speaker' => false,
                 ];
-                $currentWords = [];
+                $currentTokens = [];
                 $segmentStartMs = $startMs;
             }
 
             $currentSpeaker = $speakerTag;
-            $currentWords[] = $cleanWord;
+            $currentTokens[] = ['text' => $cleanToken, 'new_word' => (bool) $isNewWord];
         }
 
         // 마지막 세그먼트
-        if (! empty($currentWords)) {
+        if (! empty($currentTokens)) {
             $lastWord = end($words);
             $segments[] = [
                 'speaker_name' => '화자 ' . $currentSpeaker,
                 'speaker_label' => (string) $currentSpeaker,
-                'text' => $this->cleanSttText(implode(' ', $currentWords)),
+                'text' => $this->joinSentencePieceTokens($currentTokens),
                 'start_time_ms' => $segmentStartMs,
                 'end_time_ms' => $this->parseGoogleTimeToMs($lastWord['endTime'] ?? '0s'),
                 'is_manual_speaker' => false,
@@ -508,12 +512,35 @@ private function parseGoogleTimeToMs(string $timeStr): int
     }
 
     /**
-     * STT 텍스트에서 언더스코어 노이즈 제거
+     * SentencePiece 토큰 배열을 자연스러운 텍스트로 결합
+     *
+     * ▁(U+2581)가 있던 토큰은 새 단어 시작 → 앞에 공백 추가
+     * ▁가 없던 토큰은 이전 단어에 바로 붙임
+     */
+    private function joinSentencePieceTokens(array $tokens): string
+    {
+        $result = '';
+        foreach ($tokens as $i => $token) {
+            if ($i === 0) {
+                $result = $token['text'];
+            } elseif ($token['new_word']) {
+                $result .= ' ' . $token['text'];
+            } else {
+                $result .= $token['text'];
+            }
+        }
+
+        return trim(preg_replace('/\s{2,}/', ' ', $result));
+    }
+
+    /**
+     * STT 텍스트에서 SentencePiece/언더스코어 노이즈 제거
      */
     private function cleanSttText(string $text): string
     {
-        // 언더스코어 제거 후 연속 공백 정리
-        $cleaned = str_replace('_', '', $text);
+        // ▁(U+2581)를 공백으로, _(U+005F)는 제거, 연속 공백 정리
+        $cleaned = preg_replace('/\x{2581}/u', ' ', $text);
+        $cleaned = str_replace('_', '', $cleaned);
 
         return trim(preg_replace('/\s{2,}/', ' ', $cleaned));
     }
diff --git a/resources/views/juil/meeting-minutes.blade.php b/resources/views/juil/meeting-minutes.blade.php
index 3f12612a..ce34bdd6 100644
--- a/resources/views/juil/meeting-minutes.blade.php
+++ b/resources/views/juil/meeting-minutes.blade.php
@@ -947,7 +947,7 @@ className="w-full text-sm text-gray-800 leading-relaxed bg-white/70 border borde
                             )}
                         </div>
                         <div className="text-sm text-gray-800 leading-relaxed">
-                            {group.texts.map((t, ti) => <span key={ti}>{ti > 0 ? ' ' : ''}{t.text.replace(/_/g, '').replace(/\s{2,}/g, ' ').trim()}</span>)}
+                            {group.texts.map((t, ti) => <span key={ti}>{ti > 0 ? ' ' : ''}{t.text.replace(/[_\u2581]/g, ' ').replace(/\s{2,}/g, ' ').trim()}</span>)}
                         </div>
                     </div>
                 );
@@ -979,7 +979,7 @@ function ScriptView({ segments, interimText, isRecording }) {
     return (
         <div className="prose prose-sm max-w-none">
             <div className="text-sm text-gray-800 leading-relaxed whitespace-pre-wrap">
-                {segments.map((s, i) => <span key={i}>{s.text.replace(/_/g, '').replace(/\s{2,}/g, ' ').trim()}{i < segments.length - 1 ? ' ' : ''}</span>)}
+                {segments.map((s, i) => <span key={i}>{s.text.replace(/[_\u2581]/g, ' ').replace(/\s{2,}/g, ' ').trim()}{i < segments.length - 1 ? ' ' : ''}</span>)}
                 {isRecording && interimText && <span className="text-gray-400 italic"> {interimText}</span>}
             </div>
         </div>