@goodandready/dsh-tts 0.4.26 → 0.4.28

package/CHANGELOG.md

@@ -1,6 +1,45 @@
 # Changelog
 
-## 0.4.26
+## 0.4.28
+
+### Fixed
+- **Settings Contracts & Schemastery Validation** (#151, #152, #153, #155, #187, #188, #189):
+  - Made all 9 previously static fields volatile and ensured living Volatile boxes are unwrapped dynamically in live() without stale snapshots.
+  - Constrained maxQueue with .step(1).min(0) to reject non-integer and negative queue limits.
+  - Added optimistic revision validation and rollback on conflict (HTTP 409) in settings saving.
+  - Enforced isTrustedSettingsRequest auth guards on GET /dsh-tts/config and GET /dsh-tts/status.
+  - Enforced contiguous sequence ordering in SSE synthesis chunks.
+
+- **Audio Pipeline, Cache & Concurrency** (#154, #156, #157, #166, #173, #174, #175, #186, #190):
+  - Sanitized and normalized text once in prepareSpeechText() before character limits or summary generation, eliminating double-clean pronunciation distortions.
+  - Fixed subagent voice override gating with autoDetectSubagent !== false.
+  - Extended cache and dedup fingerprints with endpoint, models, voices, role, and language.
+  - Added admission concurrency throttle and semaphore for maxQueue synthesis reservations.
+  - Implemented SSE subscriber backpressure tracking and dropped slow lagging subscribers (>512 KB buffer) with drain handling.
+  - Switched cache index to an in-memory Map to eliminate disk scans on eviction.
+  - Ensured speak_text tool returns metadata and delivers audio to pending queue/SSE without dumping base64 into LLM context.
+
+- **Client Player & UI Parity** (#158, #159, #160, #161, #162, #163, #164, #165, #167, #168, #169, #176):
+  - Unified EventSource listener connection in client, preventing duplicate audio channels.
+  - Ensured skipCurrent() halts active playback immediately, and preview/replay route through the unified player.
+  - Gracefully handled browser autoplay restrictions (NotAllowedError), pausing the queue until user interaction.
+  - Kept compact recent/favorites history control visible in idle state.
+  - Registered updater routes inside ctx.effect() to properly unbind on plugin disposal.
+  - Enabled all 16 active providers in client UI, with secure key management and read-only indicators when settings scope is missing.
+  - Fixed sibling integration detection using Cordis services without relying on obsolete hasRoute.
+  - Re-anchored canonical defaults to English (en, en-US-AriaNeural) and achieved 100% parity between EN and ZH dictionaries.
+
+- **Model Weights Manager & System Hygiene** (#177, #178, #179, #180, #181, #182, #183, #184, #185):
+  - Stream chunk downloads now properly manage EventEmitter listeners without leaking error or drain handlers.
+  - Added synchronous concurrency lock per engine to share in-flight downloads and used unique nonce temporary file paths.
+  - Added minimum file size validation (minBytes / 1024 bytes) to reject truncated or error HTML downloads.
+  - Validated engine parameter in /dsh-tts/models/install and /dsh-tts/models/delete, returning HTTP 400 for unknown engines.
+  - Replaced tautological regex and source inspection tests with true runtime behavior tests.
+  - Isolated cache tests in temporary directories with automated cleanup.
+  - Synchronized README documentation with reality and strictly excluded internal dev docs (AGENTS.md, index.md, plans) from package distributions.
+  - Closes #151, #152, #153, #154, #155, #156, #157, #158, #159, #160, #161, #162, #163, #164, #165, #166, #167, #168, #169, #170, #171, #172, #173, #174, #175, #176, #177, #178, #179, #180, #181, #182, #183, #184, #185, #186, #187, #188, #189, #190, #191.
+
+## 0.4.27
 
 ### Fixed
 - **Cloud Timeout Cancellation Forwarding** (#145): Fixed unquoted `abort` identifier in `lib/providers.js:21` inside `withCloudTimeout()`. Passing an active `AbortSignal` in `synthesize()` now cleanly attaches cancellation listeners without throwing `ReferenceError`.

package/README.md

@@ -101,10 +101,10 @@ graph LR
 
 ### 4. 📚 Built-in IT Terminology Pronunciation Dictionary
 * **Pre-configured Lexicon**: Correct phonetic pronunciation for common technical abbreviations and developer terms:
-  - `SQL` $\rightarrow$ "сиквел"
-  - `Nginx` $\rightarrow$ "энджинкс"
-  - `Kubernetes` / `K8s` $\rightarrow$ "кубернетис"
-  - `Docker` $\rightarrow$ "докер", `API` $\rightarrow$ "апи", `JSON` $\rightarrow$ "джейсон", `YAML` $\rightarrow$ "ямл"
+  - `SQL` $\rightarrow$ "sequel"
+  - `Nginx` $\rightarrow$ "engine-x"
+  - `Kubernetes` / `K8s` $\rightarrow$ "coo-ber-net-ees" / "kates"
+  - `Docker` $\rightarrow$ "docker", `API` $\rightarrow$ "A-P-I", `JSON` $\rightarrow$ "jason", `YAML` $\rightarrow$ "yaml"
   - `GUI`, `CLI`, `CI/CD`, `PR`, `Regex`, `OAuth`, `HTTP`, `HTTPS`, `CPU`, `GPU`, `RAM`
 * **Interactive UI Editor**: Edit rules, preview phonetic substitutions with the **▶ Listen** button, and populate standard IT terms with one click.
 
@@ -135,7 +135,7 @@ graph LR
 | `f5` | Local F5-TTS GPU Daemon | `F5-TTS` | Default | *None* | Daemon ping only; **GPU inference not bundled** — fails honestly |
 | `elevenlabs` | ElevenLabs API | `eleven_multilingual_v2` | `Rachel` | `ELEVENLABS_API_KEY` | Ultra-realistic, emotional nuance |
 | `openai` | OpenAI Audio | `gpt-4o-mini-tts` / `tts-1` | `alloy` | `OPENAI_API_KEY` | High-quality industry standard |
-| `edge` | Microsoft Edge Online | `ru-RU-SvetlanaNeural` | `ru-RU-SvetlanaNeural` | *None* | **Free, high-fidelity neural TTS without API keys** |
+| `edge` | Microsoft Edge Online | `en-US-AriaNeural` | `en-US-AriaNeural` | *None* | **Free, high-fidelity neural TTS without API keys** |
 | `siliconflow` | SiliconFlow CosyVoice | `FunAudioLLM/CosyVoice2-0.5B` | Default | `SILICONFLOW_API_KEY` | State-of-the-art CosyVoice2 neural engine |
 | `deepinfra` | DeepInfra Kokoro | `hexgrad/Kokoro-82M` | Default | `DEEPINFRA_API_KEY` | Fast open-weights Kokoro synthesis |
 | `fireworks` | Fireworks AI | `kokoro` | Default | `FIREWORKS_API_KEY` | Ultra-low latency Kokoro inference |
@@ -225,6 +225,10 @@ Config-only (not in the card): `*KeyEnv` fields (`openaiKeyEnv`, `elevenlabsKeyE
 * `DELETE /dsh-tts/models/delete` — `{ engine: 'kokoro' | 'f5' }` → Removes local model files.
 * `GET /dsh-tts/integrations` — Status of sibling plugins (`dsh-voice`, `dsh-messenger-gateway`).
 * `GET /dsh-tts/status` — Returns active chain state, cache statistics, and engine readiness.
+* `DELETE /dsh-tts/cache` — Clears cached audio files.
+* `GET /dsh-tts/config` & `PUT /dsh-tts/config` — Retrieve and update settings revision.
+* `GET /dsh-tts/stats` & `DELETE /dsh-tts/stats` — Performance statistics and metrics reset.
+* `PUT /dsh-tts/credential` & `DELETE /dsh-tts/credential` — Securely store or remove cloud provider API keys.
 
 ---
 

package/lib/cache.js

@@ -41,56 +41,78 @@ export function createSpeechCache({ root, maxBytes }) {
     }
   }
 
-  async function listEntries() {
+  // In-memory index of disk cache entries to avoid O(N) readdir/stat/readFile per write (#190)
+  const diskEntries = new Map() // key -> { size, at }
+  let diskTotalBytes = 0
+  let indexLoaded = false
+
+  async function ensureIndex() {
     await ensure()
-    const out = []
-    for (const name of await fs.promises.readdir(dir)) {
+    if (indexLoaded) return
+    diskEntries.clear()
+    diskTotalBytes = 0
+    let files = []
+    try {
+      files = await fs.promises.readdir(dir)
+    } catch {
+      files = []
+    }
+    for (const name of files) {
       if (!name.endsWith(AUDIO_EXT)) continue
+      const key = name.slice(0, -AUDIO_EXT.length)
       const full = path.join(dir, name)
-      let st
-      try {
-        st = await fs.promises.stat(full)
-      } catch {
-        continue
-      }
-      let at = st.mtimeMs
       try {
-        const meta = JSON.parse(await fs.promises.readFile(path.join(dir, name.slice(0, -AUDIO_EXT.length) + '.json'), 'utf8'))
-        if (meta && typeof meta.at === 'number') at = meta.at
-      } catch (noMeta) { /* legacy entry without meta — evict as first */ }
-      out.push({ full, meta: name.slice(0, -AUDIO_EXT.length) + '.json', size: st.size, at })
+        const st = await fs.promises.stat(full)
+        let at = st.mtimeMs
+        try {
+          const meta = JSON.parse(await fs.promises.readFile(path.join(dir, key + '.json'), 'utf8'))
+          if (meta && typeof meta.at === 'number') at = meta.at
+        } catch { /* no meta */ }
+        diskEntries.set(key, { size: st.size, at })
+        diskTotalBytes += st.size
+      } catch { /* skip unreadable */ }
     }
-    return out
+    indexLoaded = true
   }
 
   async function evict() {
+    await ensureIndex()
     const limit = typeof maxBytes === 'function' ? maxBytes() : Number(maxBytes)
-    if (!(limit > 0)) return
-    const rawEntries = await listEntries()
-    const entries = rawEntries.map((e) => {
-      const k = path.basename(e.full, AUDIO_EXT)
-      const l1 = l1Map.get(k)
-      return { ...e, key: k, at: (l1 && typeof l1.at === 'number') ? l1.at : e.at }
-    }).sort((a, b) => a.at - b.at)
-
-    let total = entries.reduce((acc, e) => acc + e.size, 0)
-    for (const e of entries) {
-      if (total <= limit) break
-      await fs.promises.rm(e.full, { force: true })
-      await fs.promises.rm(path.join(dir, e.meta), { force: true }).catch(() => {})
-      l1Map.delete(e.key)
-      total -= e.size
+    if (!(limit > 0) || diskTotalBytes <= limit) return
+
+    // Sort in-memory entries by timestamp ascending (oldest first)
+    const sorted = Array.from(diskEntries.entries()).sort((a, b) => {
+      const atA = l1Map.get(a[0])?.at ?? a[1].at
+      const atB = l1Map.get(b[0])?.at ?? b[1].at
+      return atA - atB
+    })
+
+    for (const [key, meta] of sorted) {
+      if (diskTotalBytes <= limit) break
+      await fs.promises.rm(path.join(dir, key + AUDIO_EXT), { force: true }).catch(() => {})
+      await fs.promises.rm(path.join(dir, key + '.json'), { force: true }).catch(() => {})
+      diskEntries.delete(key)
+      diskTotalBytes -= meta.size
+      l1Map.delete(key)
     }
   }
 
   return {
     async put(key, mime, audio) {
-      await ensure()
+      await ensureIndex()
       const buf = Buffer.isBuffer(audio) ? audio : Buffer.from(audio)
       putL1(key, mime, buf)
       await fs.promises.writeFile(path.join(dir, key + AUDIO_EXT), buf)
-      await fs.promises.writeFile(path.join(dir, key + '.json'), JSON.stringify({ mime, at: Date.now() }))
-      await evict()
+      const now = Date.now()
+      await fs.promises.writeFile(path.join(dir, key + '.json'), JSON.stringify({ mime, at: now }))
+      const prev = diskEntries.get(key)
+      if (prev) diskTotalBytes -= prev.size
+      diskEntries.set(key, { size: buf.length, at: now })
+      diskTotalBytes += buf.length
+      const limit = typeof maxBytes === 'function' ? maxBytes() : Number(maxBytes)
+      if (limit > 0 && diskTotalBytes > limit) {
+        await evict()
+      }
     },
 
     async get(key) {
@@ -100,17 +122,22 @@ export function createSpeechCache({ root, maxBytes }) {
         const now = Date.now()
         l1Map.delete(key) // refresh LRU order
         l1Map.set(key, { ...hit, at: now })
+        const d = diskEntries.get(key)
+        if (d) d.at = now
         fs.promises.writeFile(path.join(dir, key + '.json'), JSON.stringify({ mime: hit.mime, at: now })).catch(() => {})
         return { mime: hit.mime || 'audio/mpeg', audio: hit.audio }
       }
 
       // 2. L2 Disk Hit
+      await ensureIndex()
       try {
         const meta = JSON.parse(await fs.promises.readFile(path.join(dir, key + '.json'), 'utf8'))
         const audio = await fs.promises.readFile(path.join(dir, key + AUDIO_EXT))
         putL1(key, meta.mime, audio)
-        // Touch LRU asynchronously so reads are never blocked.
-        fs.promises.writeFile(path.join(dir, key + '.json'), JSON.stringify({ mime: meta.mime, at: Date.now() })).catch(() => {})
+        const now = Date.now()
+        const d = diskEntries.get(key)
+        if (d) d.at = now
+        fs.promises.writeFile(path.join(dir, key + '.json'), JSON.stringify({ mime: meta.mime, at: now })).catch(() => {})
         return { mime: meta.mime || 'audio/mpeg', audio }
       } catch (miss) {
         return null
@@ -119,12 +146,15 @@ export function createSpeechCache({ root, maxBytes }) {
 
     async clear() {
       l1Map.clear()
-      const entries = await listEntries()
-      for (const e of entries) {
-        await fs.promises.rm(e.full, { force: true })
-        await fs.promises.rm(path.join(dir, e.meta), { force: true }).catch(() => {})
+      await ensureIndex()
+      const count = diskEntries.size
+      for (const key of diskEntries.keys()) {
+        await fs.promises.rm(path.join(dir, key + AUDIO_EXT), { force: true }).catch(() => {})
+        await fs.promises.rm(path.join(dir, key + '.json'), { force: true }).catch(() => {})
       }
-      return entries.length
+      diskEntries.clear()
+      diskTotalBytes = 0
+      return count
     },
   }
 }