dvgateway-adapters 1.0.0

package/dist/tts/cached-tts.d.ts

@@ -0,0 +1,131 @@
+/**
+ * Cached TTS Adapter
+ *
+ * Wraps any TtsAdapter with a disk-based audio cache keyed by
+ * (text + voiceId + provider + model). Cached audio persists across
+ * process restarts, eliminating redundant TTS API calls for repeated
+ * announcements and broadcasts.
+ *
+ * Features:
+ * - Disk-based PCM cache (survives restarts/updates)
+ * - SHA-256 cache key = hash(provider + voiceId + model + speed + text)
+ * - Automatic cache hit/miss logging with metrics
+ * - Warmup / preload API for pre-generating announcement audio pools
+ * - TTL-based expiration (optional)
+ * - Max cache size enforcement (LRU eviction)
+ *
+ * @example
+ * ```typescript
+ * import { ElevenLabsAdapter } from 'dvgateway-adapters/tts';
+ * import { CachedTtsAdapter } from 'dvgateway-adapters/tts';
+ *
+ * const tts = new CachedTtsAdapter(
+ *   new ElevenLabsAdapter({ apiKey: '...' }),
+ *   {
+ *     provider: 'elevenlabs',
+ *     cacheDir: './tts-cache',
+ *     ttlMs: 7 * 24 * 60 * 60 * 1000, // 7 days
+ *   },
+ * );
+ *
+ * // Pre-warm common announcements (done once, reused after restart)
+ * await tts.warmup([
+ *   { text: '잠시만 기다려 주세요.', voiceId: 'abc123' },
+ *   { text: '상담사에게 연결하겠습니다.' },
+ * ]);
+ *
+ * // synthesize() returns cached audio if available — no API call
+ * const audio = tts.synthesize('잠시만 기다려 주세요.');
+ * ```
+ */
+import type { TtsAdapter, TtsOptions } from 'dvgateway-sdk';
+export interface CachedTtsAdapterOptions {
+    /**
+     * Provider name used as part of the cache key.
+     * E.g. "elevenlabs", "openai"
+     */
+    provider: string;
+    /**
+     * Directory for storing cached PCM files (default: "./tts-cache")
+     * Created automatically if it does not exist.
+     */
+    cacheDir?: string;
+    /**
+     * Default voice ID used when none is specified in synthesize().
+     * Becomes part of the cache key.
+     */
+    defaultVoiceId?: string;
+    /**
+     * Default model name for cache key differentiation (e.g. "eleven_flash_v2_5").
+     */
+    defaultModel?: string;
+    /**
+     * Time-to-live for cached files in ms.
+     * 0 = no expiration (default: 0)
+     */
+    ttlMs?: number;
+    /**
+     * Maximum number of cached files.
+     * When exceeded, least-recently-used entries are evicted.
+     * 0 = unlimited (default: 0)
+     */
+    maxEntries?: number;
+}
+export interface WarmupEntry {
+    /** Text to pre-synthesize */
+    text: string;
+    /** Override voice ID for this entry */
+    voiceId?: string;
+    /** Override speed for this entry */
+    speed?: number;
+}
+interface CacheStats {
+    hits: number;
+    misses: number;
+    evictions: number;
+    totalEntries: number;
+}
+export declare class CachedTtsAdapter implements TtsAdapter {
+    private readonly inner;
+    private readonly provider;
+    private readonly cacheDir;
+    private readonly defaultVoiceId;
+    private readonly defaultModel;
+    private readonly ttlMs;
+    private readonly maxEntries;
+    private readonly stats;
+    private initialized;
+    constructor(inner: TtsAdapter, opts: CachedTtsAdapterOptions);
+    /**
+     * Synthesize text to speech, returning cached audio if available.
+     * On cache miss, delegates to the inner adapter and stores the result.
+     */
+    synthesize(text: string, options?: TtsOptions): AsyncIterable<Buffer>;
+    /**
+     * Pre-generate and cache audio for a list of announcements.
+     * Skips entries that are already cached. Useful at startup to
+     * build an audio pool that survives restarts.
+     *
+     * @returns Number of entries that were newly synthesized
+     */
+    warmup(entries: ReadonlyArray<WarmupEntry>): Promise<number>;
+    /**
+     * Check if a specific text is already cached.
+     */
+    isCached(text: string, voiceId?: string, speed?: number): Promise<boolean>;
+    /**
+     * Clear all cached audio files.
+     */
+    clearCache(): Promise<number>;
+    /**
+     * Get cache statistics.
+     */
+    getStats(): Readonly<CacheStats>;
+    private buildCacheKey;
+    private ensureDir;
+    private readCache;
+    private writeCache;
+    private evictIfNeeded;
+}
+export {};
+//# sourceMappingURL=cached-tts.d.ts.map

package/dist/tts/cached-tts.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cached-tts.d.ts","sourceRoot":"","sources":["../../src/tts/cached-tts.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAuCG;AAKH,OAAO,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAI5D,MAAM,WAAW,uBAAuB;IACtC;;;OAGG;IACH,QAAQ,EAAE,MAAM,CAAC;IAEjB;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAC;IAElB;;;OAGG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;IAExB;;OAEG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IAEtB;;;OAGG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IAEf;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,WAAW;IAC1B,6BAA6B;IAC7B,IAAI,EAAE,MAAM,CAAC;IACb,uCAAuC;IACvC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,oCAAoC;IACpC,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,UAAU,UAAU;IAClB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;CACtB;AASD,qBAAa,gBAAiB,YAAW,UAAU;IACjD,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAa;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAS;IACxC,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAS;IACtC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAS;IAC/B,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAS;IACpC,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAqE;IAC3F,OAAO,CAAC,WAAW,CAAS;gBAEhB,KAAK,EAAE,UAAU,EAAE,IAAI,EAAE,uBAAuB;IAU5D;;;OAGG;IACI,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,UAAU,GAAG,aAAa,CAAC,MAAM,CAAC;IAsC5E;;;;;;OAMG;IACG,MAAM,CAAC,OAAO,EAAE,aAAa,CAAC,WAAW,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC;IA+BlE;;OAEG;IACG,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,MAAM,EAAE,KAAK,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAQhF;;OAEG;IACG,UAAU,IAAI,OAAO,CAAC,MAAM,CAAC;IAanC;;OAEG;IACH,QAAQ,IAAI,QAAQ,CAAC,UAAU,CAAC;IAMhC,OAAO,CAAC,aAAa;YAKP,SAAS;YAMT,SAAS;YAkBT,UAAU;YAUV,aAAa;CAgC5B"}

package/dist/tts/cached-tts.js

@@ -0,0 +1,231 @@
+/**
+ * Cached TTS Adapter
+ *
+ * Wraps any TtsAdapter with a disk-based audio cache keyed by
+ * (text + voiceId + provider + model). Cached audio persists across
+ * process restarts, eliminating redundant TTS API calls for repeated
+ * announcements and broadcasts.
+ *
+ * Features:
+ * - Disk-based PCM cache (survives restarts/updates)
+ * - SHA-256 cache key = hash(provider + voiceId + model + speed + text)
+ * - Automatic cache hit/miss logging with metrics
+ * - Warmup / preload API for pre-generating announcement audio pools
+ * - TTL-based expiration (optional)
+ * - Max cache size enforcement (LRU eviction)
+ *
+ * @example
+ * ```typescript
+ * import { ElevenLabsAdapter } from 'dvgateway-adapters/tts';
+ * import { CachedTtsAdapter } from 'dvgateway-adapters/tts';
+ *
+ * const tts = new CachedTtsAdapter(
+ *   new ElevenLabsAdapter({ apiKey: '...' }),
+ *   {
+ *     provider: 'elevenlabs',
+ *     cacheDir: './tts-cache',
+ *     ttlMs: 7 * 24 * 60 * 60 * 1000, // 7 days
+ *   },
+ * );
+ *
+ * // Pre-warm common announcements (done once, reused after restart)
+ * await tts.warmup([
+ *   { text: '잠시만 기다려 주세요.', voiceId: 'abc123' },
+ *   { text: '상담사에게 연결하겠습니다.' },
+ * ]);
+ *
+ * // synthesize() returns cached audio if available — no API call
+ * const audio = tts.synthesize('잠시만 기다려 주세요.');
+ * ```
+ */
+import { createHash } from 'node:crypto';
+import { mkdir, readFile, writeFile, readdir, stat, unlink } from 'node:fs/promises';
+import { join } from 'node:path';
+// ─── Constants ───────────────────────────────────────────────────────────────
+const DEFAULT_CACHE_DIR = './tts-cache';
+const PCM_CHUNK_SIZE = 640; // 20ms at 16kHz, 16-bit = 640 bytes
+// ─── Implementation ──────────────────────────────────────────────────────────
+export class CachedTtsAdapter {
+    inner;
+    provider;
+    cacheDir;
+    defaultVoiceId;
+    defaultModel;
+    ttlMs;
+    maxEntries;
+    stats = { hits: 0, misses: 0, evictions: 0, totalEntries: 0 };
+    initialized = false;
+    constructor(inner, opts) {
+        this.inner = inner;
+        this.provider = opts.provider;
+        this.cacheDir = opts.cacheDir ?? DEFAULT_CACHE_DIR;
+        this.defaultVoiceId = opts.defaultVoiceId ?? 'default';
+        this.defaultModel = opts.defaultModel ?? 'default';
+        this.ttlMs = opts.ttlMs ?? 0;
+        this.maxEntries = opts.maxEntries ?? 0;
+    }
+    /**
+     * Synthesize text to speech, returning cached audio if available.
+     * On cache miss, delegates to the inner adapter and stores the result.
+     */
+    async *synthesize(text, options) {
+        await this.ensureDir();
+        const voiceId = options?.voiceId ?? this.defaultVoiceId;
+        const speed = options?.speed ?? 1.0;
+        const key = this.buildCacheKey(text, voiceId, speed);
+        const filePath = join(this.cacheDir, `${key}.pcm`);
+        // Try cache hit
+        const cached = await this.readCache(filePath);
+        if (cached) {
+            this.stats.hits++;
+            // Yield in 20ms chunks to match real-time streaming cadence
+            let offset = 0;
+            while (offset < cached.length) {
+                const end = Math.min(offset + PCM_CHUNK_SIZE, cached.length);
+                yield cached.subarray(offset, end);
+                offset = end;
+            }
+            return;
+        }
+        // Cache miss — synthesize via inner adapter
+        this.stats.misses++;
+        const chunks = [];
+        for await (const chunk of this.inner.synthesize(text, options)) {
+            chunks.push(chunk);
+            yield chunk;
+        }
+        // Store in cache asynchronously (don't block the caller)
+        const full = Buffer.concat(chunks);
+        this.writeCache(filePath, full).catch(() => {
+            // Swallow write errors — cache is best-effort
+        });
+    }
+    /**
+     * Pre-generate and cache audio for a list of announcements.
+     * Skips entries that are already cached. Useful at startup to
+     * build an audio pool that survives restarts.
+     *
+     * @returns Number of entries that were newly synthesized
+     */
+    async warmup(entries) {
+        await this.ensureDir();
+        let synthesized = 0;
+        for (const entry of entries) {
+            const voiceId = entry.voiceId ?? this.defaultVoiceId;
+            const speed = entry.speed ?? 1.0;
+            const key = this.buildCacheKey(entry.text, voiceId, speed);
+            const filePath = join(this.cacheDir, `${key}.pcm`);
+            // Skip if already cached and not expired
+            const existing = await this.readCache(filePath);
+            if (existing) {
+                continue;
+            }
+            // Synthesize and store
+            const chunks = [];
+            const options = { voiceId, speed };
+            for await (const chunk of this.inner.synthesize(entry.text, options)) {
+                chunks.push(chunk);
+            }
+            const full = Buffer.concat(chunks);
+            await this.writeCache(filePath, full);
+            synthesized++;
+        }
+        return synthesized;
+    }
+    /**
+     * Check if a specific text is already cached.
+     */
+    async isCached(text, voiceId, speed) {
+        await this.ensureDir();
+        const key = this.buildCacheKey(text, voiceId ?? this.defaultVoiceId, speed ?? 1.0);
+        const filePath = join(this.cacheDir, `${key}.pcm`);
+        const cached = await this.readCache(filePath);
+        return cached !== undefined;
+    }
+    /**
+     * Clear all cached audio files.
+     */
+    async clearCache() {
+        await this.ensureDir();
+        const files = await readdir(this.cacheDir);
+        const pcmFiles = files.filter(f => f.endsWith('.pcm'));
+        let removed = 0;
+        for (const f of pcmFiles) {
+            await unlink(join(this.cacheDir, f)).catch(() => undefined);
+            removed++;
+        }
+        this.stats.totalEntries = 0;
+        return removed;
+    }
+    /**
+     * Get cache statistics.
+     */
+    getStats() {
+        return { ...this.stats };
+    }
+    // ─── Private ─────────────────────────────────────────────────────────────
+    buildCacheKey(text, voiceId, speed) {
+        const raw = `${this.provider}|${this.defaultModel}|${voiceId}|${speed}|${text}`;
+        return createHash('sha256').update(raw).digest('hex');
+    }
+    async ensureDir() {
+        if (this.initialized)
+            return;
+        await mkdir(this.cacheDir, { recursive: true });
+        this.initialized = true;
+    }
+    async readCache(filePath) {
+        try {
+            // Check TTL
+            if (this.ttlMs > 0) {
+                const info = await stat(filePath);
+                const age = Date.now() - info.mtimeMs;
+                if (age > this.ttlMs) {
+                    await unlink(filePath).catch(() => undefined);
+                    return undefined;
+                }
+            }
+            return await readFile(filePath);
+        }
+        catch {
+            return undefined;
+        }
+    }
+    async writeCache(filePath, data) {
+        // Enforce max entries via LRU eviction before writing
+        if (this.maxEntries > 0) {
+            await this.evictIfNeeded();
+        }
+        await writeFile(filePath, data);
+        this.stats.totalEntries++;
+    }
+    async evictIfNeeded() {
+        try {
+            const files = await readdir(this.cacheDir);
+            const pcmFiles = files.filter(f => f.endsWith('.pcm'));
+            if (pcmFiles.length < this.maxEntries)
+                return;
+            // Sort by modification time (oldest first) for LRU eviction
+            const withStats = await Promise.all(pcmFiles.map(async (f) => {
+                const fullPath = join(this.cacheDir, f);
+                const info = await stat(fullPath);
+                return { path: fullPath, mtimeMs: info.mtimeMs };
+            }));
+            withStats.sort((a, b) => a.mtimeMs - b.mtimeMs);
+            // Remove oldest entries until we're under the limit
+            const toRemove = withStats.length - this.maxEntries + 1;
+            for (let i = 0; i < toRemove; i++) {
+                const entry = withStats[i];
+                if (entry) {
+                    await unlink(entry.path).catch(() => undefined);
+                    this.stats.evictions++;
+                    this.stats.totalEntries--;
+                }
+            }
+        }
+        catch {
+            // Best-effort eviction
+        }
+    }
+}
+//# sourceMappingURL=cached-tts.js.map

package/dist/tts/cached-tts.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cached-tts.js","sourceRoot":"","sources":["../../src/tts/cached-tts.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAuCG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AACrF,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AA2DjC,gFAAgF;AAEhF,MAAM,iBAAiB,GAAG,aAAa,CAAC;AACxC,MAAM,cAAc,GAAG,GAAG,CAAC,CAAC,oCAAoC;AAEhE,gFAAgF;AAEhF,MAAM,OAAO,gBAAgB;IACV,KAAK,CAAa;IAClB,QAAQ,CAAS;IACjB,QAAQ,CAAS;IACjB,cAAc,CAAS;IACvB,YAAY,CAAS;IACrB,KAAK,CAAS;IACd,UAAU,CAAS;IACnB,KAAK,GAAe,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,SAAS,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAE,CAAC;IACnF,WAAW,GAAG,KAAK,CAAC;IAE5B,YAAY,KAAiB,EAAE,IAA6B;QAC1D,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;QACnB,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC;QAC9B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,iBAAiB,CAAC;QACnD,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC,cAAc,IAAI,SAAS,CAAC;QACvD,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,SAAS,CAAC;QACnD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;QAC7B,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC;IACzC,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,CAAC,UAAU,CAAC,IAAY,EAAE,OAAoB;QAClD,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;QAEvB,MAAM,OAAO,GAAG,OAAO,EAAE,OAAO,IAAI,IAAI,CAAC,cAAc,CAAC;QACxD,MAAM,KAAK,GAAG,OAAO,EAAE,KAAK,IAAI,GAAG,CAAC;QACpC,MAAM,GAAG,GAAG,IAAI,CAAC,aAAa,CAAC,IAAI,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;QACrD,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,GAAG,MAAM,CAAC,CAAC;QAEnD,gBAAgB;QAChB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;QAC9C,IAAI,MAAM,EAAE,CAAC;YACX,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC;YAClB,4DAA4D;YAC5D,IAAI,MAAM,GAAG,CAAC,CAAC;YACf,OAAO,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC;gBAC9B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,GAAG,cAAc,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;gBAC7D,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;gBACnC,MAAM,GAAG,GAAG,CAAC;YACf,CAAC;YACD,OAAO;QACT,CAAC;QAED,4CAA4C;QAC5C,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;QACpB,MAAM,MAAM,GAAa,EAAE,CAAC;QAE5B,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,IAAI,EAAE,OAAO,CAAC,EAAE,CAAC;YAC/D,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACnB,MAAM,KAAK,CAAC;QACd,CAAC;QAED,yDAAyD;QACzD,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACnC,IAAI,CAAC,UAAU,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE;YACzC,8CAA8C;QAChD,CAAC,CAAC,CAAC;IACL,CAAC;IAED;;;;;;OAMG;IACH,KAAK,CAAC,MAAM,CAAC,OAAmC;QAC9C,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;QAEvB,IAAI,WAAW,GAAG,CAAC,CAAC;QAEpB,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,MAAM,OAAO,GAAG,KAAK,CAAC,OAAO,IAAI,IAAI,CAAC,cAAc,CAAC;YACrD,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,IAAI,GAAG,CAAC;YACjC,MAAM,GAAG,GAAG,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;YAC3D,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,GAAG,MAAM,CAAC,CAAC;YAEnD,yCAAyC;YACzC,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;YAChD,IAAI,QAAQ,EAAE,CAAC;gBACb,SAAS;YACX,CAAC;YAED,uBAAuB;YACvB,MAAM,MAAM,GAAa,EAAE,CAAC;YAC5B,MAAM,OAAO,GAAe,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;YAC/C,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,IAAI,CAAC,KAAK,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,EAAE,OAAO,CAAC,EAAE,CAAC;gBACrE,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YACrB,CAAC;YACD,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YACnC,MAAM,IAAI,CAAC,UAAU,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;YACtC,WAAW,EAAE,CAAC;QAChB,CAAC;QAED,OAAO,WAAW,CAAC;IACrB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,QAAQ,CAAC,IAAY,EAAE,OAAgB,EAAE,KAAc;QAC3D,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;QACvB,MAAM,GAAG,GAAG,IAAI,CAAC,aAAa,CAAC,IAAI,EAAE,OAAO,IAAI,IAAI,CAAC,cAAc,EAAE,KAAK,IAAI,GAAG,CAAC,CAAC;QACnF,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,GAAG,MAAM,CAAC,CAAC;QACnD,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC;QAC9C,OAAO,MAAM,KAAK,SAAS,CAAC;IAC9B,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,UAAU;QACd,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;QACvB,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC3C,MAAM,QAAQ,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;QACvD,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;YACzB,MAAM,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;YAC5D,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,YAAY,GAAG,CAAC,CAAC;QAC5B,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,QAAQ;QACN,OAAO,EAAE,GAAG,IAAI,CAAC,KAAK,EAAE,CAAC;IAC3B,CAAC;IAED,4EAA4E;IAEpE,aAAa,CAAC,IAAY,EAAE,OAAe,EAAE,KAAa;QAChE,MAAM,GAAG,GAAG,GAAG,IAAI,CAAC,QAAQ,IAAI,IAAI,CAAC,YAAY,IAAI,OAAO,IAAI,KAAK,IAAI,IAAI,EAAE,CAAC;QAChF,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACxD,CAAC;IAEO,KAAK,CAAC,SAAS;QACrB,IAAI,IAAI,CAAC,WAAW;YAAE,OAAO;QAC7B,MAAM,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAChD,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;IAC1B,CAAC;IAEO,KAAK,CAAC,SAAS,CAAC,QAAgB;QACtC,IAAI,CAAC;YACH,YAAY;YACZ,IAAI,IAAI,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;gBACnB,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,CAAC;gBAClC,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,OAAO,CAAC;gBACtC,IAAI,GAAG,GAAG,IAAI,CAAC,KAAK,EAAE,CAAC;oBACrB,MAAM,MAAM,CAAC,QAAQ,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;oBAC9C,OAAO,SAAS,CAAC;gBACnB,CAAC;YACH,CAAC;YAED,OAAO,MAAM,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAClC,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,UAAU,CAAC,QAAgB,EAAE,IAAY;QACrD,sDAAsD;QACtD,IAAI,IAAI,CAAC,UAAU,GAAG,CAAC,EAAE,CAAC;YACxB,MAAM,IAAI,CAAC,aAAa,EAAE,CAAC;QAC7B,CAAC;QAED,MAAM,SAAS,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QAChC,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,CAAC;IAC5B,CAAC;IAEO,KAAK,CAAC,aAAa;QACzB,IAAI,CAAC;YACH,MAAM,KAAK,GAAG,MAAM,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;YAC3C,MAAM,QAAQ,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC;YAEvD,IAAI,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,UAAU;gBAAE,OAAO;YAE9C,4DAA4D;YAC5D,MAAM,SAAS,GAAG,MAAM,OAAO,CAAC,GAAG,CACjC,QAAQ,CAAC,GAAG,CAAC,KAAK,EAAC,CAAC,EAAC,EAAE;gBACrB,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;gBACxC,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,CAAC;gBAClC,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;YACnD,CAAC,CAAC,CACH,CAAC;YAEF,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC;YAEhD,oDAAoD;YACpD,MAAM,QAAQ,GAAG,SAAS,CAAC,MAAM,GAAG,IAAI,CAAC,UAAU,GAAG,CAAC,CAAC;YACxD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,EAAE,CAAC,EAAE,EAAE,CAAC;gBAClC,MAAM,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;gBAC3B,IAAI,KAAK,EAAE,CAAC;oBACV,MAAM,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAC;oBAChD,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;oBACvB,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,CAAC;gBAC5B,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;IACH,CAAC;CACF"}

package/dist/tts/elevenlabs.d.ts

@@ -0,0 +1,95 @@
+/**
+ * ElevenLabs TTS Adapter
+ *
+ * Synthesizes text to speech using ElevenLabs' streaming API.
+ * Returns 16kHz slin16 PCM chunks for direct injection into DVGateway.
+ *
+ * Features:
+ * - Streaming output (first audio chunk ≤75ms latency with Flash models)
+ * - Automatic 24kHz → 16kHz resampling (ElevenLabs outputs 24kHz PCM)
+ * - Turbo / Flash / Multilingual model support
+ * - Korean + multilingual support
+ * - Voice settings: stability, similarity boost, style, speed
+ *
+ * ElevenLabs Model Reference (2026-03):
+ *   eleven_flash_v2_5      — Fastest (~75ms TTFA), best for real-time voice (default)
+ *   eleven_turbo_v2_5      — Balanced quality/speed (~200ms TTFA)
+ *   eleven_multilingual_v2 — Highest multilingual quality (higher latency)
+ *   eleven_multilingual_v3 — Next-gen multilingual, improved prosody (2026)
+ *   eleven_english_sts_v2  — English speech-to-speech transformation
+ *
+ * Voice IDs (popular Korean/multilingual voices):
+ *   21m00Tcm4TlvDq8ikWAM  — Rachel (English, warm)
+ *   29vD33N1CtxCmqQRPOHJ  — Drew (English, conversational)
+ *   pNInz6obpgDQGcFmaJgB  — Adam (English, deep)
+ *   Use ElevenLabs voice library or clone a custom voice for Korean.
+ *
+ * API Endpoint: POST https://api.elevenlabs.io/v1/text-to-speech/{voiceId}/stream
+ * Docs: https://elevenlabs.io/docs/api-reference/text-to-speech
+ */
+import type { TtsAdapter, TtsOptions, HumanVoiceOptions, VoiceInfo } from 'dvgateway-sdk';
+export interface ElevenLabsAdapterOptions {
+    apiKey: string;
+    /** Voice ID (default: "Rachel" voice — specify a Korean-capable voice for KO) */
+    voiceId?: string;
+    /**
+     * Model ID (default: "eleven_multilingual_v2" when humanVoice enabled, "eleven_flash_v2_5" otherwise)
+     * Options:
+     *   eleven_flash_v2_5      — Fastest (~75ms), best for real-time voice
+     *   eleven_turbo_v2_5      — Balanced quality/speed
+     *   eleven_multilingual_v2 — Best multilingual quality (higher latency)
+     *   eleven_multilingual_v3 — Next-gen multilingual (2026, improved prosody)
+     */
+    model?: string;
+    /** Stability (0.0–1.0, default: 0.3 with humanVoice, 0.5 otherwise) — higher = more consistent voice */
+    stability?: number;
+    /** Similarity boost (0.0–1.0, default: 0.75) — higher = closer to original voice */
+    similarityBoost?: number;
+    /** Style exaggeration (0.0–1.0, default: 0.6 with humanVoice, 0.0 otherwise) — adds expressiveness, increases latency */
+    style?: number;
+    /** Boost speaker clarity and target speaker similarity (default: true) */
+    useSpeakerBoost?: boolean;
+    /**
+     * Output format (default: "pcm_24000")
+     * Options: pcm_16000, pcm_22050, pcm_24000, pcm_44100, mp3_44100_128, etc.
+     * Use pcm_24000 for best quality streaming; we resample to 16kHz for DVGateway.
+     */
+    outputFormat?: string;
+    /**
+     * Optimize streaming latency (0–4, default: 4 — maximum optimization)
+     * 0 = off (best quality), 4 = maximum latency optimization
+     * When humanVoice is enabled, defaults to 3 for better prosody.
+     */
+    optimizeStreamingLatency?: number;
+    /**
+     * Human-like voice optimization (default: Korean-optimized preset).
+     * When enabled, adjusts stability, style, and model for natural speech.
+     * Set to false to disable, or provide custom HumanVoiceOptions.
+     */
+    humanVoice?: HumanVoiceOptions | false;
+}
+/** Built-in Korean native voices from ElevenLabs Voice Library */
+export declare const ELEVENLABS_KOREAN_VOICES: ReadonlyArray<VoiceInfo>;
+export declare class ElevenLabsAdapter implements TtsAdapter {
+    private readonly opts;
+    constructor(opts: ElevenLabsAdapterOptions);
+    synthesize(text: string, opts?: TtsOptions): AsyncIterable<Buffer>;
+    /**
+     * Fetch all voices available to the user from ElevenLabs API.
+     * Returns default voices, cloned voices, and shared library voices.
+     * Cloned voices are labeled with (클론), generated with (생성됨).
+     */
+    static fetchVoices(apiKey: string): Promise<VoiceInfo[]>;
+    /**
+     * Clone a voice from an audio file using ElevenLabs API.
+     *
+     * @param apiKey - ElevenLabs API key
+     * @param name - Name for the cloned voice
+     * @param audioData - Raw audio file (Buffer/Uint8Array)
+     * @param fileName - Original file name for MIME type detection
+     * @param description - Optional voice description
+     * @returns VoiceInfo with the new voice's id and name
+     */
+    static cloneVoice(apiKey: string, name: string, audioData: Uint8Array, fileName?: string, description?: string): Promise<VoiceInfo>;
+}
+//# sourceMappingURL=elevenlabs.d.ts.map

package/dist/tts/elevenlabs.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"elevenlabs.d.ts","sourceRoot":"","sources":["../../src/tts/elevenlabs.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA4BG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AAG1F,MAAM,WAAW,wBAAwB;IACvC,MAAM,EAAE,MAAM,CAAC;IACf,iFAAiF;IACjF,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB;;;;;;;OAOG;IACH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,wGAAwG;IACxG,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,oFAAoF;IACpF,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,yHAAyH;IACzH,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,0EAA0E;IAC1E,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;;;;OAIG;IACH,wBAAwB,CAAC,EAAE,MAAM,CAAC;IAClC;;;;OAIG;IACH,UAAU,CAAC,EAAE,iBAAiB,GAAG,KAAK,CAAC;CACxC;AAKD,kEAAkE;AAClE,eAAO,MAAM,wBAAwB,EAAE,aAAa,CAAC,SAAS,CAUpD,CAAC;AAEX,qBAAa,iBAAkB,YAAW,UAAU;IAClD,OAAO,CAAC,QAAQ,CAAC,IAAI,CAA+G;gBAExH,IAAI,EAAE,wBAAwB;IA4BnC,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,UAAU,GAAG,aAAa,CAAC,MAAM,CAAC;IAwEzE;;;;OAIG;WACU,WAAW,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,SAAS,EAAE,CAAC;IA6B9D;;;;;;;;;OASG;WACU,UAAU,CACrB,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,SAAS,EAAE,UAAU,EACrB,QAAQ,GAAE,MAAoB,EAC9B,WAAW,GAAE,MAAW,GACvB,OAAO,CAAC,SAAS,CAAC;CAoBtB"}

package/dist/tts/elevenlabs.js

@@ -0,0 +1,195 @@
+/**
+ * ElevenLabs TTS Adapter
+ *
+ * Synthesizes text to speech using ElevenLabs' streaming API.
+ * Returns 16kHz slin16 PCM chunks for direct injection into DVGateway.
+ *
+ * Features:
+ * - Streaming output (first audio chunk ≤75ms latency with Flash models)
+ * - Automatic 24kHz → 16kHz resampling (ElevenLabs outputs 24kHz PCM)
+ * - Turbo / Flash / Multilingual model support
+ * - Korean + multilingual support
+ * - Voice settings: stability, similarity boost, style, speed
+ *
+ * ElevenLabs Model Reference (2026-03):
+ *   eleven_flash_v2_5      — Fastest (~75ms TTFA), best for real-time voice (default)
+ *   eleven_turbo_v2_5      — Balanced quality/speed (~200ms TTFA)
+ *   eleven_multilingual_v2 — Highest multilingual quality (higher latency)
+ *   eleven_multilingual_v3 — Next-gen multilingual, improved prosody (2026)
+ *   eleven_english_sts_v2  — English speech-to-speech transformation
+ *
+ * Voice IDs (popular Korean/multilingual voices):
+ *   21m00Tcm4TlvDq8ikWAM  — Rachel (English, warm)
+ *   29vD33N1CtxCmqQRPOHJ  — Drew (English, conversational)
+ *   pNInz6obpgDQGcFmaJgB  — Adam (English, deep)
+ *   Use ElevenLabs voice library or clone a custom voice for Korean.
+ *
+ * API Endpoint: POST https://api.elevenlabs.io/v1/text-to-speech/{voiceId}/stream
+ * Docs: https://elevenlabs.io/docs/api-reference/text-to-speech
+ */
+import { resample, float32ToSlin16, HUMAN_VOICE_DEFAULTS_KO } from 'dvgateway-sdk';
+const ELEVENLABS_SAMPLE_RATE = 24000;
+const DV_SAMPLE_RATE = 16000;
+/** Built-in Korean native voices from ElevenLabs Voice Library */
+export const ELEVENLABS_KOREAN_VOICES = [
+    { id: 'pjJMvFj0JGWi3mogOkHH', label: 'Hyun Bin (남성, 한국어)' },
+    { id: 't0jbNlBVZ17f02VDIeMI', label: '지영 / JiYoung (여성, 한국어)' },
+    { id: 'zrHiDhphv9ZnVXBqCLjz', label: 'Jennie (여성, 한국어)' },
+    { id: 'ZJCNdOEhQGMOIbMuhBME', label: 'Han Aim (남성, 한국어)' },
+    { id: 'ova4yY2jqnnUdGOmTGbx', label: 'KKC HQ (남성, 한국어)' },
+    { id: 'Xb7hH8MSUJpSbSDYk0k2', label: 'Anna Kim (여성, 한국어)' },
+    { id: 'XrExE9yKIg1WjnnlVkGX', label: 'Yuna (여성, 한국어)' },
+    { id: 'ThT5KcBeYPX3keUQqHPh', label: 'Jina (여성, 한국어)' },
+    { id: 'Sita5M0jWFxPiECPABjR', label: 'jjeong (여성, 한국어)' },
+];
+export class ElevenLabsAdapter {
+    opts;
+    constructor(opts) {
+        // Resolve human voice options: default is Korean-optimized preset
+        const hv = opts.humanVoice === false
+            ? false
+            : { ...HUMAN_VOICE_DEFAULTS_KO, ...(opts.humanVoice ?? {}) };
+        // When humanVoice is enabled, adjust defaults for natural speech:
+        //   - model: eleven_multilingual_v2 (best Korean prosody)
+        //   - stability: lower (0.3) for more natural variation
+        //   - style: higher (emotionalRange) for expressiveness
+        //   - optimizeStreamingLatency: 3 (balance quality & speed)
+        const hvEnabled = hv !== false;
+        this.opts = {
+            voiceId: opts.voiceId ?? '21m00Tcm4TlvDq8ikWAM', // Rachel
+            model: opts.model ?? (hvEnabled ? 'eleven_multilingual_v2' : 'eleven_flash_v2_5'),
+            stability: opts.stability ?? (hvEnabled ? (1.0 - hv.speechVariation) : 0.5),
+            similarityBoost: opts.similarityBoost ?? 0.75,
+            style: opts.style ?? (hvEnabled ? hv.emotionalRange : 0.0),
+            useSpeakerBoost: opts.useSpeakerBoost ?? true,
+            outputFormat: opts.outputFormat ?? 'pcm_24000',
+            optimizeStreamingLatency: opts.optimizeStreamingLatency ?? (hvEnabled ? 3 : 4),
+            apiKey: opts.apiKey,
+            humanVoice: hv,
+        };
+    }
+    async *synthesize(text, opts) {
+        const voiceId = opts?.voiceId ?? this.opts.voiceId;
+        const url = `https://api.elevenlabs.io/v1/text-to-speech/${voiceId}/stream`
+            + `?output_format=${this.opts.outputFormat}`
+            + `&optimize_streaming_latency=${this.opts.optimizeStreamingLatency}`;
+        const response = await fetch(url, {
+            method: 'POST',
+            headers: {
+                'xi-api-key': this.opts.apiKey,
+                'Content-Type': 'application/json',
+                Accept: 'audio/pcm',
+            },
+            body: JSON.stringify({
+                text,
+                model_id: this.opts.model,
+                voice_settings: {
+                    stability: this.opts.stability,
+                    similarity_boost: this.opts.similarityBoost,
+                    style: this.opts.style,
+                    use_speaker_boost: this.opts.useSpeakerBoost,
+                    speed: opts?.speed ?? 1.0,
+                },
+            }),
+        });
+        if (!response.ok) {
+            const errBody = await response.text().catch(() => '');
+            throw new Error(`ElevenLabs TTS failed (HTTP ${response.status}): ${errBody}`);
+        }
+        if (!response.body) {
+            throw new Error('ElevenLabs TTS: empty response body');
+        }
+        // Stream PCM audio chunks through resampler
+        const reader = response.body.getReader();
+        let remainder = Buffer.alloc(0);
+        // PCM 24kHz frames: process in 960-byte chunks (20ms at 24kHz, 16-bit = 960 bytes)
+        const CHUNK_BYTES = 960; // 480 samples × 2 bytes
+        while (true) {
+            const { done, value } = await reader.read();
+            if (done) {
+                // Flush any remaining bytes
+                if (remainder.length > 0) {
+                    const samples24k = pcmBytesToFloat32(remainder);
+                    const samples16k = resample(samples24k, ELEVENLABS_SAMPLE_RATE, DV_SAMPLE_RATE);
+                    yield float32ToSlin16(samples16k);
+                }
+                break;
+            }
+            // Accumulate into buffer
+            remainder = Buffer.concat([remainder, Buffer.from(value)]);
+            // Emit complete 20ms frames
+            while (remainder.length >= CHUNK_BYTES) {
+                const frame = remainder.subarray(0, CHUNK_BYTES);
+                remainder = remainder.subarray(CHUNK_BYTES);
+                const samples24k = pcmBytesToFloat32(frame);
+                const samples16k = resample(samples24k, ELEVENLABS_SAMPLE_RATE, DV_SAMPLE_RATE);
+                yield float32ToSlin16(samples16k);
+            }
+        }
+    }
+    // ── Voice Management APIs ──────────────────────────────────────────────
+    /**
+     * Fetch all voices available to the user from ElevenLabs API.
+     * Returns default voices, cloned voices, and shared library voices.
+     * Cloned voices are labeled with (클론), generated with (생성됨).
+     */
+    static async fetchVoices(apiKey) {
+        const response = await fetch('https://api.elevenlabs.io/v1/voices', {
+            headers: { 'xi-api-key': apiKey },
+        });
+        if (!response.ok) {
+            const errBody = await response.text().catch(() => '');
+            throw new Error(`ElevenLabs fetch voices failed (HTTP ${response.status}): ${errBody}`);
+        }
+        const data = await response.json();
+        return data.voices.map((v) => {
+            let tag = '';
+            if (v.category === 'cloned')
+                tag = ' (클론)';
+            else if (v.category === 'generated')
+                tag = ' (생성됨)';
+            else if (v.category === 'professional')
+                tag = ' (프로)';
+            const lang = v.labels?.language ? ` [${v.labels.language}]` : '';
+            return { id: v.voice_id, label: `${v.name}${tag}${lang}` };
+        });
+    }
+    /**
+     * Clone a voice from an audio file using ElevenLabs API.
+     *
+     * @param apiKey - ElevenLabs API key
+     * @param name - Name for the cloned voice
+     * @param audioData - Raw audio file (Buffer/Uint8Array)
+     * @param fileName - Original file name for MIME type detection
+     * @param description - Optional voice description
+     * @returns VoiceInfo with the new voice's id and name
+     */
+    static async cloneVoice(apiKey, name, audioData, fileName = 'voice.wav', description = '') {
+        const formData = new FormData();
+        formData.append('name', name);
+        if (description)
+            formData.append('description', description);
+        formData.append('files', new Blob([audioData]), fileName);
+        const response = await fetch('https://api.elevenlabs.io/v1/voices/add', {
+            method: 'POST',
+            headers: { 'xi-api-key': apiKey },
+            body: formData,
+        });
+        if (!response.ok) {
+            const errBody = await response.text().catch(() => '');
+            throw new Error(`ElevenLabs voice clone failed (HTTP ${response.status}): ${errBody}`);
+        }
+        const data = await response.json();
+        return { id: data.voice_id, label: data.name };
+    }
+}
+/** Convert little-endian 16-bit PCM bytes to Float32Array */
+function pcmBytesToFloat32(buf) {
+    const samples = buf.byteLength >> 1;
+    const out = new Float32Array(samples);
+    for (let i = 0; i < samples; i++) {
+        out[i] = buf.readInt16LE(i * 2) / 32768.0;
+    }
+    return out;
+}
+//# sourceMappingURL=elevenlabs.js.map