@drakulavich/kesha-voice-kit 1.22.0 → 1.23.0

package/NOTICES.md

@@ -33,6 +33,16 @@ Text-to-speech (Russian, 5 baked-in speakers). Apache 2.0. Model: [alphacep/vosk
 
 **Runtime code:** Originally pulled from `crates.io` as `vosk-tts-rs 0.1.0`. As of #216 the runtime subset is **vendored** under `rust/vendor/vosk-tts/` (Apache 2.0; copied from [`andreytkachenko/vosk-tts-rs`](https://github.com/andreytkachenko/vosk-tts-rs) — see `rust/vendor/vosk-tts/LICENSE` and `rust/vendor/vosk-tts/NOTICE`). The vendored copy drops the upstream gRPC server/CLI/HTTP-model-fetch modules and replaces the `tokenizers` crate with a small inline BERT WordPiece tokenizer, eliminating the `prost`, `tonic`, `tokio`, `axum`, `hyper`, `reqwest`, `tokenizers`, `onig_sys`, `esaxx-rs`, `bzip2-sys`, `lzma-sys`, and `zstd-sys` transitive dependencies that broke Windows MSVC linkage.
 
+### klebster — CharsiuG2P byt5-tiny ONNX (multilingual G2P)
+
+Grapheme-to-phoneme for es/fr/it/pt Kokoro voices (`kesha install --tts`, #212).
+CC-BY 4.0. ONNX export by [klebster](https://huggingface.co/klebster/g2p_multilingual_byT5_tiny_onnx)
+from the original CharsiuG2P model.
+
+**Original model:** Zhu et al. 2022, "Multilingual Grapheme-to-Phoneme Conversion
+with Language Neutral and Language Specific Representations",
+[arXiv:2204.03067](https://arxiv.org/abs/2204.03067).
+
 ### misaki-rs (English G2P)
 
 Embedded grapheme-to-phoneme for English (Kokoro pipeline, #207). MIT. Source: [`misaki-rs`](https://github.com/MicheleYin/misaki-rs).

package/man/kesha.1

@@ -1,4 +1,4 @@
-.TH KESHA 1 "May 2026" "kesha" "User Commands"
+.TH KESHA 1 "2026" "kesha" "User Commands"
 .SH NAME
 kesha \- open-source local voice toolkit
 .SH SYNOPSIS

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@drakulavich/kesha-voice-kit",
-  "version": "1.22.0",
+  "version": "1.23.0",
   "description": "Give your tools a voice — speech to text and back, 25 languages, up to ~19× faster than Whisper. On your machine.",
   "type": "module",
   "bin": {
@@ -29,10 +29,16 @@
   "openclaw": {
     "extensions": [
       "./openclaw-plugin.cjs"
-    ]
+    ],
+    "compat": {
+      "pluginApi": ">=2026.5.27"
+    },
+    "build": {
+      "openclawVersion": "2026.5.27"
+    }
   },
   "keshaEngine": {
-    "version": "1.22.0"
+    "version": "1.23.0"
   },
   "scripts": {
     "test": "bun test",

package/src/install-plan.ts

@@ -62,9 +62,24 @@ const VAD_FILES: PlanFile[] = [
   { relPath: "models/silero-vad/silero_vad.onnx", sizeBytes: 2_327_524 },
 ];
 
+// klebster CharsiuG2P byt5-tiny ONNX export (CC-BY 4.0 — see NOTICES.md).
+// 3 files enabling multilingual G2P for es/fr/it/pt Kokoro voices (#212).
+const G2P_CHARSIU_FILES: PlanFile[] = [
+  { relPath: "models/g2p/byt5-tiny/encoder_model.onnx", sizeBytes: 12_478_704 },
+  { relPath: "models/g2p/byt5-tiny/decoder_model.onnx", sizeBytes: 11_983_268 },
+  { relPath: "models/g2p/byt5-tiny/decoder_with_past_model.onnx", sizeBytes: 5_427_260 },
+];
+
 const KOKORO_FILES: PlanFile[] = [
   { relPath: "models/kokoro-82m/model.onnx", sizeBytes: 325_532_387 },
   { relPath: "models/kokoro-82m/voices/am_michael.bin", sizeBytes: 522_240 },
+  // Multilingual voice packs for es/fr/it/pt (#212).
+  // em_alex (es, male), ff_siwis (fr, female — only French voice in Kokoro v1.0),
+  // im_nicola (it, male), pm_alex (pt, male).
+  { relPath: "models/kokoro-82m/voices/em_alex.bin", sizeBytes: 522_240 },
+  { relPath: "models/kokoro-82m/voices/ff_siwis.bin", sizeBytes: 522_240 },
+  { relPath: "models/kokoro-82m/voices/im_nicola.bin", sizeBytes: 522_240 },
+  { relPath: "models/kokoro-82m/voices/pm_alex.bin", sizeBytes: 522_240 },
 ];
 
 const VOSK_RU_FILES: PlanFile[] = [
@@ -245,11 +260,21 @@ export async function renderInstallPlan(options: InstallPlanOptions = {}): Promi
       components.push(
         bundleComponent(
           cacheRoot,
-          "TTS Kokoro EN",
+          "TTS Kokoro EN/ES/FR/IT/PT",
           "model cache",
           KOKORO_FILES,
           noCache,
-          "English en-* voices",
+          "English en-* voices + multilingual es-*/fr-*/it-*/pt-* voices",
+        ),
+      );
+      components.push(
+        bundleComponent(
+          cacheRoot,
+          "G2P CharsiuG2P byt5-tiny",
+          "model cache",
+          G2P_CHARSIU_FILES,
+          noCache,
+          "multilingual grapheme-to-phoneme for es/fr/it/pt Kokoro voices (CC-BY 4.0)",
         ),
       );
       components.push(

package/src/shell-artifacts.ts

@@ -257,9 +257,11 @@ function roff(value: string): string {
 }
 
 function manpageDate(): string {
-  const now = new Date();
-  const month = now.toLocaleString("en-US", { month: "long" });
-  return `${month} ${now.getFullYear()}`;
+  // Year only — the man `.TH` date drove a monthly drift failure in the
+  // shell-artifacts test (committed "May 2026" vs regenerated "June 2026").
+  // Year granularity is plenty for a manpage revision date and stays stable
+  // across the year.
+  return `${new Date().getFullYear()}`;
 }
 
 function renderManpage(model: ArtifactModel): string {

package/src/voice-routing.ts

@@ -14,6 +14,19 @@ const DARWIN_KOKORO_DEFAULTS: Record<string, string> = {
   zh: "zh-zm_yunjian",
 };
 
+/**
+ * ONNX-platform (Linux / Windows / Intel macOS) multilingual defaults.
+ * These mirror the Rust `default_voice_for_lang` in the engine.
+ * es/it/pt are male; fr is the documented brand-rule exception
+ * (Kokoro v1.0 ships no male French voice).
+ */
+const ONNX_KOKORO_DEFAULTS: Record<string, string> = {
+  es: "es-em_alex",
+  fr: "fr-ff_siwis",
+  it: "it-im_nicola",
+  pt: "pt-pm_alex",
+};
+
 /** Map a detected language code to a default voice id. Unknown / low-confidence → undefined. */
 export function pickVoiceForLang(
   code: string | undefined,
@@ -30,10 +43,11 @@ export function pickVoiceForLang(
       // AVSpeech Milena is a macOS system voice (any arch); Vosk elsewhere.
       return platform === "darwin" ? RU_DARWIN_FALLBACK_VOICE : "ru-vosk-m02";
     default:
-      // Multilingual Kokoro voices ship only on darwin-arm64 (FluidAudio ANE).
-      // Intel Macs have no such voice pack, so don't auto-route to ids the
-      // engine can't honour — fall through to the engine default instead.
+      // darwin-arm64: FluidAudio ANE voice pack (full multilingual set).
       if (platform === "darwin" && arch === "arm64") return DARWIN_KOKORO_DEFAULTS[baseCode];
-      return undefined;
+      // ONNX platforms (Linux, Windows, Intel macOS): es/fr/it/pt are supported
+      // via CharsiuG2P + ONNX Kokoro. Other languages (hi, ja, zh, …) have no
+      // ONNX voice pack, so fall through to the engine default (undefined).
+      return ONNX_KOKORO_DEFAULTS[baseCode];
   }
 }