@drakulavich/kesha-voice-kit 1.17.1 → 1.18.0

package/README.md

@@ -100,7 +100,7 @@ Each segment gets a `speaker` integer (cluster ID, stable within one file). Linu
 
 ## Text-to-speech
 
-Kesha speaks back via Kokoro-82M (English) and Vosk-TTS (Russian) — voice auto-picks from the text's language:
+Kesha speaks back via Kokoro-82M (English) and Vosk-TTS (Russian) — voice auto-picks from the text's language. On darwin-arm64, Kokoro uses FluidAudio CoreML instead of ONNX:
 
 ```bash
 kesha install --tts                      # ~990MB (Kokoro + Vosk-TTS RU, opt-in)
@@ -158,7 +158,7 @@ See [BENCHMARK.md](BENCHMARK.md) for the full per-file breakdown (Russian + Engl
 | SpeechBrain ECAPA-TDNN | Audio language detection | ~86MB | [HuggingFace](https://huggingface.co/speechbrain/lang-id-voxlingua107-ecapa) |
 | Apple NLLanguageRecognizer | Text language detection | built-in | macOS system framework |
 | Silero VAD v5 (opt-in) | Voice activity detection | ~2.3MB | [snakers4/silero-vad](https://github.com/snakers4/silero-vad) |
-| Kokoro-82M / Vosk-TTS (opt-in) | Text-to-speech | ~990MB | [Kokoro](https://huggingface.co/hexgrad/Kokoro-82M) · [Vosk-TTS](https://github.com/alphacep/vosk-tts) |
+| Kokoro-82M / Vosk-TTS (opt-in) | Text-to-speech | ~990MB | [FluidAudio Kokoro](https://github.com/FluidInference/FluidAudio) on darwin-arm64; ONNX Kokoro elsewhere · [Vosk-TTS](https://github.com/alphacep/vosk-tts) |
 
 All models run through `kesha-engine` — a Rust binary using [FluidAudio](https://github.com/FluidInference/FluidAudio) (CoreML) on Apple Silicon and [ort](https://github.com/pykeio/ort) (ONNX Runtime) on other platforms.
 

package/SKILL.md

@@ -22,7 +22,7 @@ Local voice toolkit: transcribe voice messages to text, synthesize speech, detec
 ## When to use
 
 - **Voice memo arrived** (Telegram, WhatsApp, Slack, Signal .ogg/.opus/.m4a): transcribe with `kesha --json <path>` and branch on the detected language.
-- **Need to reply with audio (file playback)**: synthesize with `kesha say "<text>" > reply.wav`. Auto-routes by detected language (Kokoro-82M for English, Vosk-TTS for Russian). For other languages and ~180 more voices use `--voice macos-*` on macOS (zero model download).
+- **Need to reply with audio (file playback)**: synthesize with `kesha say "<text>" > reply.wav`. Auto-routes by detected language (Kokoro-82M for English, Vosk-TTS for Russian). On darwin-arm64, English Kokoro uses FluidAudio CoreML instead of ONNX. For other languages and ~180 more voices use `--voice macos-*` on macOS (zero model download).
 - **Need to send a voice note (Telegram, WhatsApp, Signal, Discord)**: synthesize directly into the messenger-native format with `kesha say "<text>" --format ogg-opus --out reply.ogg`. Default is mono 24 kHz @ 32 kbps — what Telegram `sendVoice` expects. No `ffmpeg` round-trip needed.
 - **Need to detect what language a file is in** before choosing a pipeline: `kesha --json audio.ogg` returns both audio-based and text-based language detection with confidence scores.
 
@@ -153,7 +153,7 @@ No system deps — English G2P is embedded (`misaki-rs`); Russian G2P is bundled
 
 **Speech-to-text (25):** Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Russian, Slovak, Slovenian, Spanish, Swedish, Ukrainian.
 
-**Text-to-speech:** English (Kokoro-82M, ~70 voices), Russian (Vosk-TTS, 5 baked-in speakers — default `ru-vosk-m02`), plus any macOS system voice via `--voice macos-*`.
+**Text-to-speech:** English (Kokoro-82M; FluidAudio CoreML on darwin-arm64, ONNX elsewhere), Russian (Vosk-TTS, 5 baked-in speakers — default `ru-vosk-m02`), plus any macOS system voice via `--voice macos-*`.
 
 ## Performance
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@drakulavich/kesha-voice-kit",
-  "version": "1.17.1",
+  "version": "1.18.0",
   "description": "Open-source voice toolkit. Speech-to-text (25 langs, ~19x faster than Whisper on Apple Silicon, ONNX fallback on CPU), text-to-speech (Kokoro + Vosk-TTS + ~180 macOS system voices, SSML), voice-activity detection, language detection (107 langs). Rust engine, OpenClaw skill.",
   "type": "module",
   "bin": {
@@ -30,7 +30,7 @@
     ]
   },
   "keshaEngine": {
-    "version": "1.17.0"
+    "version": "1.18.0"
   },
   "scripts": {
     "test": "bun test",

package/src/engine-install.ts

@@ -1,5 +1,6 @@
 import { dirname, join } from "path";
-import { existsSync, mkdirSync, chmodSync, accessSync, constants } from "fs";
+import { tmpdir } from "os";
+import { existsSync, mkdirSync, chmodSync, accessSync, constants, rmSync } from "fs";
 import {
   getEngineBinPath,
   getEngineCapabilities,
@@ -74,6 +75,13 @@ const SIDECARS: SidecarSpec[] = [
     availableHint: "--speakers available",
     unavailableHint: "--speakers unavailable",
   },
+  {
+    fileBasename: "kesha-kokoro",
+    assetName: "kesha-kokoro-darwin-arm64",
+    displayName: "FluidAudio Kokoro sidecar",
+    availableHint: "Kokoro uses FluidAudio CoreML",
+    unavailableHint: "en-* Kokoro voices unavailable on Darwin",
+  },
   {
     // Runtime resolver looks for plain `kesha-textlang` next to the engine
     // (see `rust/src/text_lang.rs::helper_path`), not the platform-suffixed
@@ -221,6 +229,76 @@ async function downloadSidecar(
   }
 }
 
+async function warmDarwinKokoro(binPath: string): Promise<void> {
+  if (process.platform !== "darwin" || process.arch !== "arm64") return;
+  const sidecarPath = join(dirname(binPath), "kesha-kokoro");
+  if (!existsSync(sidecarPath)) return;
+
+  const outPath = join(tmpdir(), `kesha-kokoro-warmup-${process.pid}.wav`);
+  log.progress("Warming FluidAudio Kokoro CoreML cache...");
+
+  const startedAt = performance.now();
+  const proc = Bun.spawn(
+    [
+      binPath,
+      "say",
+      "--voice",
+      "en-am_michael",
+      "--out",
+      outPath,
+      "Kesha warmup.",
+    ],
+    {
+      stdout: "ignore",
+      stderr: "pipe",
+    },
+  );
+
+  let timedOut = false;
+  const timer = setTimeout(() => {
+    timedOut = true;
+    proc.kill();
+  }, 180_000);
+
+  let stderr = "";
+  try {
+    const stderrStream = proc.stderr as ReadableStream<Uint8Array>;
+    const [stderrText, exitCode] = await Promise.all([
+      new Response(stderrStream).text(),
+      proc.exited,
+    ]);
+    stderr = stderrText.trim();
+
+    if (timedOut) {
+      log.warn("FluidAudio Kokoro warmup timed out; first `kesha say en-*` may still be slow.");
+      return;
+    }
+    if (exitCode !== 0) {
+      log.warn(
+        `FluidAudio Kokoro warmup failed${stderr ? `: ${stderr}` : ""}; first ` +
+          "`kesha say en-*` may still be slow.",
+      );
+      return;
+    }
+
+    log.success(
+      `FluidAudio Kokoro warmed (${Math.round(performance.now() - startedAt)}ms).`,
+    );
+  } catch (e) {
+    log.warn(
+      `FluidAudio Kokoro warmup failed (${e instanceof Error ? e.message : e}); ` +
+        "first `kesha say en-*` may still be slow.",
+    );
+  } finally {
+    clearTimeout(timer);
+    try {
+      rmSync(outPath, { force: true });
+    } catch {
+      // best-effort cleanup only
+    }
+  }
+}
+
 export interface InstallOptions {
   /** Also install Kokoro + Vosk-TTS models. */
   tts?: boolean;
@@ -427,6 +505,10 @@ export async function downloadEngine(
     throw new Error(detail ? `Failed to install models: ${detail}` : "Failed to install models");
   }
 
+  if (options.tts) {
+    await warmDarwinKokoro(binPath);
+  }
+
   log.success("Backend installed successfully.");
   return binPath;
 }