Speed up TTS playback by default

.env.example

@@ -14,6 +14,7 @@ TTS_LANGUAGE=KR
 TTS_SPEAKER=KR
 TTS_DEVICE=cpu
 TTS_SPEED=1.18
+TTS_PLAYBACK_RATE=3
 TTS_SDP_RATIO=0.22
 TTS_NOISE_SCALE=0.55
 TTS_NOISE_SCALE_W=0.75

README.md

@@ -85,6 +85,9 @@ bun run test:tts -- "안녕하세요. 로컬 티티에스 테스트입니다."
   - Docker GPU passthrough를 쓸 때만 `cuda`로 바꿉니다
 - `TTS_SPEED`
   - 기본값 `1.18`
+- `TTS_PLAYBACK_RATE`
+  - 기본값 `3`
+  - 생성된 WAV를 `ffmpeg`로 더 빠르게 재생합니다
 - `TTS_SDP_RATIO`
   - 기본값 `0.22`
 - `TTS_NOISE_SCALE`
@@ -187,6 +190,7 @@ TTS_LANGUAGE=KR
 TTS_SPEAKER=KR
 TTS_DEVICE=cpu
 TTS_SPEED=1.18
+TTS_PLAYBACK_RATE=3
 TTS_SDP_RATIO=0.22
 TTS_NOISE_SCALE=0.55
 TTS_NOISE_SCALE_W=0.75

src/config.ts

@@ -25,6 +25,7 @@ const envSchema = z.object({
   TTS_SPEAKER: z.string().min(1).default("KR"),
   TTS_DEVICE: z.string().min(1).default("cpu"),
   TTS_SPEED: z.coerce.number().min(0.5).max(2).default(1.18),
+  TTS_PLAYBACK_RATE: z.coerce.number().min(0.5).max(4).default(3),
   TTS_SDP_RATIO: z.coerce.number().min(0).max(1).default(0.22),
   TTS_NOISE_SCALE: z.coerce.number().min(0).max(2).default(0.55),
   TTS_NOISE_SCALE_W: z.coerce.number().min(0).max(2).default(0.75),

src/services/audio-playback.ts

@@ -1,5 +1,9 @@
 import { spawn } from "node:child_process";
+import { rm } from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
 import process from "node:process";
+import { randomUUID } from "node:crypto";
 
 async function run(command: string, args: string[], env?: NodeJS.ProcessEnv): Promise<void> {
   await new Promise<void>((resolve, reject) => {
@@ -20,7 +24,48 @@ async function run(command: string, args: string[], env?: NodeJS.ProcessEnv): Pr
   });
 }
 
-export async function playWavFile(filePath: string): Promise<void> {
+function buildAtempoFilter(rate: number): string {
+  const filters: string[] = [];
+  let remaining = rate;
+
+  while (remaining > 2) {
+    filters.push("atempo=2.0");
+    remaining /= 2;
+  }
+
+  while (remaining < 0.5) {
+    filters.push("atempo=0.5");
+    remaining /= 0.5;
+  }
+
+  filters.push(`atempo=${remaining.toFixed(3)}`);
+  return filters.join(",");
+}
+
+async function applyPlaybackRate(filePath: string, playbackRate: number): Promise<string> {
+  if (Math.abs(playbackRate - 1) < 0.01) {
+    return filePath;
+  }
+
+  const targetPath = path.join(os.tmpdir(), `realtime-voice-bot-tts-${randomUUID()}.wav`);
+  const filter = buildAtempoFilter(playbackRate);
+
+  await run("ffmpeg", [
+    "-y",
+    "-hide_banner",
+    "-loglevel",
+    "error",
+    "-i",
+    filePath,
+    "-filter:a",
+    filter,
+    targetPath,
+  ]);
+
+  return targetPath;
+}
+
+async function playNativeWavFile(filePath: string): Promise<void> {
   if (process.platform === "win32") {
     const env = {
       ...process.env,
@@ -45,3 +90,15 @@ export async function playWavFile(filePath: string): Promise<void> {
 
   throw new Error(`지원하지 않는 플랫폼입니다: ${process.platform}`);
 }
+
+export async function playWavFile(filePath: string, playbackRate = 1): Promise<void> {
+  const playablePath = await applyPlaybackRate(filePath, playbackRate);
+
+  try {
+    await playNativeWavFile(playablePath);
+  } finally {
+    if (playablePath !== filePath) {
+      await rm(playablePath, { force: true }).catch(() => undefined);
+    }
+  }
+}

src/services/melo-tts.ts

@@ -67,7 +67,7 @@ export class MeloTtsService {
 
     try {
       await this.synthesizeToFile(trimmed, targetPath);
-      await playWavFile(targetPath);
+      await playWavFile(targetPath, this.config.TTS_PLAYBACK_RATE);
     } finally {
       await rm(targetPath, { force: true }).catch(() => undefined);
     }
@@ -128,6 +128,7 @@ export class MeloTtsService {
       language: this.config.TTS_LANGUAGE,
       speaker: this.config.TTS_SPEAKER,
       speed: this.config.TTS_SPEED,
+      playback_rate: this.config.TTS_PLAYBACK_RATE,
       sdp_ratio: this.config.TTS_SDP_RATIO,
       noise_scale: this.config.TTS_NOISE_SCALE,
       noise_scale_w: this.config.TTS_NOISE_SCALE_W,