dsh-agent-voice 0.1.5 → 0.1.6

package/lib/client.js

@@ -186,30 +186,32 @@ window.__ModuleLoader__.load({
         }, []);
 
         function playAnnouncement(p) {
-          // Audio is served via HTTP (settings stays small). audioPath is the
-          // absolute file path; the route is the public /dsh-agent-voice/audio/<name>.
-          let url = p.audioUrl;
-          if (!url && p.audioPath) {
-            const name = String(p.audioPath).split("/").pop();
-            url = "/dsh-agent-voice/audio/" + name;
-          }
-          const audio = new Audio(url);
-          audio.volume = Math.max(0, Math.min(1, Number(p.volume ?? 1)));
-          const kindLabel = p.kind === "attention" ? "🔔 Pozor: " : (p.kind === "task" ? "✅ " : "🔊 ");
-          setToast(kindLabel + p.text);
-          if (toastTimer.current) clearTimeout(toastTimer.current);
-          toastTimer.current = setTimeout(() => setToast(null), 6000);
-          audio.play().catch((err) => {
-            // Autoplay still blocked: keep the toast and retry on next gesture.
-            pendingRef.current = p;
-          });
+          // The WAV is fetched one-shot via the action round-trip (base64),
+          // then played. Settings never carries the audio bytes.
+          const doPlay = (dataUrl) => {
+            const audio = new Audio(dataUrl);
+            audio.volume = Math.max(0, Math.min(1, Number(p.volume ?? 1)));
+            const kindLabel = p.kind === "attention" ? "🔔 Pozor: " : (p.kind === "task" ? "✅ " : "🔊 ");
+            setToast(kindLabel + p.text);
+            if (toastTimer.current) clearTimeout(toastTimer.current);
+            toastTimer.current = setTimeout(() => setToast(null), 6000);
+            audio.play().catch(() => {
+              // Autoplay still blocked: keep the toast and retry on next gesture.
+              pendingRef.current = p;
+            });
+          };
+          if (p.dataUrl) { doPlay(p.dataUrl); return; }
+          // One-shot poll for the audio bytes by announcement id.
+          action({ action: "poll", id: p.id }).then((r) => {
+            if (r && r.ok && r.result && r.result.found && r.result.dataUrl) doPlay(r.result.dataUrl);
+          }).catch(() => {});
         }
 
         React.useEffect(() => {
           if (!statusJson) return;
           let p = null;
           try { p = JSON.parse(statusJson); } catch { return; }
-          if (!p || (!p.dataUrl && !p.audioPath && !p.audioUrl) || !p.id) return;
+          if (!p || !p.id) return;
           if (p.id === lastPlayed) return;
           setLastPlayed(p.id);
           playAnnouncement(p);
@@ -237,7 +239,7 @@ window.__ModuleLoader__.load({
         id: "agent-voice-player",
         order: 100,
         label: () => "🔊 Agent Voice přehrávač",
-        inject: () => ({ config, status }),
+        inject: () => ({ action, config, status }),
         children: {},
       }, Player));
 

package/lib/index.js

@@ -48,6 +48,10 @@ const DEFAULTS = {
 
 const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
 
+// One-shot audio cache keyed by announcement id: the client polls action=poll
+// to fetch the WAV bytes (base64) for a given id, then the entry expires.
+const audioCache = new Map();
+
 let workerProcess = null;
 
 async function ensureWorker() {
@@ -97,30 +101,6 @@ async function postWorker(path, body) {
 }
 
 function apply(ctx) {
-  // Serve generated audio to the client over HTTP (keeps WAVs out of the
-  // settings document entirely).
-  const ws = ctx.get("webServer");
-  if (ws) {
-    ws.register({
-      kind: "prefix",
-      path: "/dsh-agent-voice/audio/",
-      handler: async (req, res) => {
-        const name = decodeURIComponent(req.url.split("/").pop() || "");
-        if (!/^av-\d+-[a-z0-9]+\.wav$/.test(name)) {
-          res.writeHead(404).end("not found");
-          return;
-        }
-        try {
-          const data = await readFile(join(AUDIO_DIR, name));
-          res.writeHead(200, { "Content-Type": "audio/wav", "Content-Length": data.length });
-          res.end(data);
-        } catch {
-          res.writeHead(404).end("not found");
-        }
-      },
-    });
-  }
-
   let source = () => ({ ...DEFAULTS });
   installSettingsSection(ctx, NS, SCHEMA, DEFAULTS, {
     setSource: (getCurrent) => { source = () => ({ ...DEFAULTS, ...(getCurrent() ?? {}) }); },
@@ -135,24 +115,20 @@ function apply(ctx) {
     response: z.string(),
   }), { base: {} });
 
-  // Audio files are served to the client over a webServer route instead of
-  // being base64-inlined into the settings document. A full WAV in settings
-  // would bloat the shared file by hundreds of KB and stall every settings
-  // round-trip (which is exactly what broke the UI before this fix).
+  // Audio bytes travel to the client via the ACTION round-trip (one-shot,
+  // never persisted): the client calls action=poll with an announcement id and
+  // the host returns the base64 WAV. The STATUS namespace carries only small
+  // metadata (id + text), so the shared settings document stays tiny.
   async function synthesize(text, voice, language) {
     await ensureWorker();
     const res = await postWorker("/tts", { text, voice, language });
-    await mkdir(AUDIO_DIR, { recursive: true });
-    const name = `av-${Date.now()}-${Math.random().toString(36).slice(2, 8)}.wav`;
-    const file = join(AUDIO_DIR, name);
-    await writeFile(file, await readFile(res.audio));
-    return file;
+    return await readFile(res.audio);
   }
 
   async function announce(kind, text) {
     const cfg = source();
     if (!cfg.enabled) return { ok: false, skipped: "disabled" };
-    const file = await synthesize(text, cfg.voice, cfg.language);
+    const wav = await synthesize(text, cfg.voice, cfg.language);
     const id = String(Date.now()) + "-" + Math.random().toString(36).slice(2);
     const payload = {
       id,
@@ -160,24 +136,14 @@ function apply(ctx) {
       text,
       voice: cfg.voice,
       volume: cfg.volume,
-      audioPath: file, // the client fetches /dsh-agent-voice/audio/<name>
       at: Date.now(),
     };
     await ctx.settings.replace(STATUS_NS, { json: JSON.stringify(payload) });
-    cleanupOldAudio().catch(() => {});
+    audioCache.set(id, wav);
+    setTimeout(() => audioCache.delete(id), 60000).unref?.();
     return { ok: true, kind, text };
   }
 
-  async function cleanupOldAudio() {
-    const { readdir, rm } = await import("node:fs/promises");
-    let files = [];
-    try { files = (await readdir(AUDIO_DIR)).filter((f) => f.endsWith(".wav")).sort(); } catch { return; }
-    while (files.length > 10) {
-      const old = files.shift();
-      try { await rm(join(AUDIO_DIR, old), { force: true }); } catch {}
-    }
-  }
-
   async function runAction(req) {
     const cfg = source();
     const a = req.action;
@@ -214,6 +180,14 @@ function apply(ctx) {
       const text = req.text || "Ahoj, tady agent. Test hlasu proběhl úspěšně.";
       return announce("test", text);
     }
+    if (a === "poll") {
+      // One-shot audio fetch: the client asks for the WAV of an announcement
+      // id and receives it as base64 (never persisted to settings).
+      if (!req.id) throw new Error("poll requires id");
+      const wav = audioCache.get(req.id);
+      if (!wav) return { id: req.id, found: false };
+      return { id: req.id, found: true, dataUrl: "data:audio/wav;base64," + Buffer.from(wav).toString("base64") };
+    }
     throw new Error("unknown action: " + a);
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-agent-voice",
-  "version": "0.1.5",
+  "version": "0.1.6",
   "description": "DeepSeek Harness plugin: gives the agent a voice — it reads completed tasks and attention requests aloud (never reasoning or interim reports) via local Piper / XTTS v2, with a settings panel and a curated set of voices.",
   "type": "module",
   "main": "lib/index.js",