Render custom councils 5% slower

A multi-voice custom debate is denser than a single-figure mode, more
voices, more interruption, more arguments to follow. The slower pace
gives the listener room. Applied at the TTS request layer so the
gateway adjusts speed via ffmpeg atempo and pitch stays correct;
Web Audio playbackRate would have shifted pitch about one semitone.

Author: michael-chipmates

client/src/services/council/CustomCouncilService.ts

@@ -19,6 +19,14 @@ import { getAudioContext } from '../audio/audioQueueManager';
 import { splitForGatewayCap } from '../audio/utils/splitForGatewayCap';
 import { sendConversion, COUNCIL_ENGAGED_THRESHOLD_S } from '../../utils/public/gclidCapture';
 
+// Custom councils render 5% slower than the user's general TTS speed. The
+// debate is denser than a single-figure mode (more voices, more interruption,
+// more arguments to follow), so the slower pace gives the listener room.
+// Applied at the TTS request layer so the gateway adjusts speed correctly
+// (ffmpeg atempo preserves pitch); doing it via Web Audio playbackRate would
+// shift pitch ~1 semitone.
+const COUNCIL_SPEED_MULTIPLIER = 0.95;
+
 // ============================================
 // Type Definitions
 // ============================================
@@ -1033,11 +1041,12 @@ export class CustomCouncilService {
 
       // Per-chunk filename suffix prevents server-side cache collisions when
       // the same segment-id maps to multiple TTS calls.
+      const councilSpeed = ttsSettings.speed * COUNCIL_SPEED_MULTIPLIER;
       const renderChunk = (chunkText: string, idx: number): Promise<AudioFile> => {
         const chunkBaseName = chunks.length > 1 ? `${sessionPrefixedId}_c${idx + 1}` : sessionPrefixedId;
         return explicitVoice
-          ? this._generateTTSWithExplicitVoice(chunkText, chunkBaseName, explicitVoice, ttsService, ttsSettings.speed, language, sessionId)
-          : convertTextToSpeech(chunkText, chunkBaseName, speakerId, ttsService, ttsSettings.speed, language, sessionId);
+          ? this._generateTTSWithExplicitVoice(chunkText, chunkBaseName, explicitVoice, ttsService, councilSpeed, language, sessionId)
+          : convertTextToSpeech(chunkText, chunkBaseName, speakerId, ttsService, councilSpeed, language, sessionId);
       };
 
       // Parallel render for multi-chunk segments — same speaker means same