@drakulavich/kesha-voice-kit 1.28.0 → 1.29.1

package/README.md

@@ -36,14 +36,14 @@ bun add -g @drakulavich/kesha-voice-kit
 kesha --version                                 # confirms `kesha` resolved on PATH
 
 # 3. Download the engine and models — pick one path
-kesha init                                      # guided: backend, languages, TTS voices
+kesha init                                      # guided: TTS languages and optional VAD / diarization
 kesha install --plan && kesha install           # manual: preview the sizes, then download
 
 # 4. Transcribe
 kesha audio.ogg                                 # transcript to stdout
 ```
 
-`kesha install` pulls ~2.5 GB on Linux/Windows and ~0.6 GB on Apple Silicon, whose CoreML engine reads a smaller model set. It is always explicit — nothing downloads behind your back — and the model step has no progress bar, so expect a few quiet minutes. If `bun --version` fails right after step 1, reload your PATH: `exec $SHELL -l`.
+`kesha install` pulls ~2.5 GB on Linux/Windows and ~0.6 GB on Apple Silicon, whose CoreML engine reads a smaller model set. It is always explicit — nothing downloads behind your back — and reports download progress on stderr. If `bun --version` fails right after step 1, reload your PATH: `exec $SHELL -l`.
 
 Prefer Homebrew or Docker? See [Other install methods](#other-install-methods).
 Air-gapped or behind a corporate mirror? See [docs/model-mirror.md](docs/model-mirror.md).
@@ -90,15 +90,16 @@ $ kesha freedom.ogg tahiti.ogg
 ```
 
 - **Record from the mic (macOS):** `kesha record --out hello.wav` writes microphone audio to a WAV file (`kesha hello.wav` transcribes it). macOS prompts for microphone access on first use — grant it under System Settings → Privacy & Security → Microphone if it was denied. On Linux/Windows or headless boxes, pass any existing audio file straight to `kesha` instead.
-- **Dictate straight to text (darwin-arm64):** `kesha record --live` transcribes the mic as it captures and prints the transcript to stdout — no WAV in between, so it pipes (`kesha record --live | pbcopy`). Progress goes to stderr. Other platforms keep the two-step `record --out` + transcribe flow. An interruption is recoverable: Ctrl-C (or SIGTERM) stops the session, still prints what you dictated, and exits 130/143, and the audio is spilled to a recovery WAV under `~/.cache/kesha/recordings/` — named on stderr when the session starts, deleted once the transcript has actually been delivered, kept if anything — a signal, a crash, a closed terminal, a dead pipe — got in the way first ([#962](https://github.com/drakulavich/kesha-voice-kit/issues/962)).
+- **Dictate straight to text (darwin-arm64):** `kesha record --live` transcribes the mic as it captures and prints the transcript to stdout — no WAV in between, so it pipes (`kesha record --live | pbcopy`). To end after trailing silence, explicitly install VAD then opt in: `kesha install --vad && kesha record --live --auto-stop`. The defaults are 1,000 ms of silence after 250 ms of speech; tune them with `--auto-stop-silence-ms`, `--auto-stop-min-speech-ms`, and `--auto-stop-threshold`. Progress goes to stderr. Linux and Windows do not capture the microphone; pass an existing audio file to `kesha` to transcribe it. An interruption is recoverable: Ctrl-C (or SIGTERM) stops the session, still prints what you dictated, and exits 130/143, and the audio is spilled to a recovery WAV under `~/.cache/kesha/recordings/` — named on stderr when the session starts, deleted once the transcript has actually been delivered, kept if anything — a signal, a crash, a closed terminal, a dead pipe — got in the way first ([#962](https://github.com/drakulavich/kesha-voice-kit/issues/962)).
 - **Long / silence-heavy audio:** install VAD (`kesha install --vad`); Kesha auto-uses it past 120 s. Without VAD, long audio falls back to fixed ASR chunks. See [docs/vad.md](docs/vad.md).
 - **Speaker diarization** (darwin-arm64): `kesha install --diarize` (which installs VAD too), then `kesha --json --speakers meeting.m4a` stamps each segment with a `speaker` id. `--speakers` engages VAD windowing itself at any duration, so it cannot be combined with `--no-vad`. Linux/Windows return a clear "darwin-arm64 only" error ([#199](https://github.com/drakulavich/kesha-voice-kit/issues/199)).
 - **Word-level timestamps** (every platform): `kesha --json --timestamps audio.ogg` adds a `words` array to each segment — `{ "word": "email", "start": 0.72, "end": 1.12 }` — on the same file-relative clock as the segment, so a word always lies inside the segment carrying it. Read them off the decoder's own frame grid, so: times are quantised to 0.08 s, consecutive spans may overlap (each `end` is a per-word duration prediction, not the next word's `start`), `end >= start` rather than strictly greater, and punctuation stays attached to its word. The key is simply absent where a segment has none — any segment `--itn` rewrote, for one — so check the `transcribe.words` capability rather than expecting an empty array ([#720](https://github.com/drakulavich/kesha-voice-kit/issues/720)).
+- **Text-language detection:** JSON and TOON results include `textLanguage` with a language code, confidence, and its `source`. On macOS Kesha uses Apple `NLLanguageRecognizer`; elsewhere it uses the bundled `tinyld` fallback, whose confidence scale is different. This is separate from `audioLanguage`, which identifies the spoken audio when available.
 - **Written-form numbers:** `--itn` rewrites what the model spells out — `"two hundred thirty two"` → `"232"`, `"five dollars and fifty cents"` → `"$5.50"`. Opt-in, every platform, timestamps untouched. English-only in practice; Russian and the rest pass through unchanged. Spoken punctuation names stay words (`"dot"`, `"comma"`, `"the period of growth"`) because Kesha transcribes speech rather than dictation — so `"example dot com"` keeps its words too ([#822](https://github.com/drakulavich/kesha-voice-kit/issues/822)). A sentence `"and"` survives the number that follows it (`"cats and three dogs"` → `"cats and 3 dogs"`), while an `"and"` the number owns still joins it (`"three hundred and five"` → `"305"`) ([#1000](https://github.com/drakulavich/kesha-voice-kit/issues/1000)) — and no longer splits the number around it (`"two hundred and thirty two"` → `"232"`, not `"230 2"`) ([#1006](https://github.com/drakulavich/kesha-voice-kit/issues/1006)). A hyphenated number reads the same as the spaced form (`"twenty-five apples"` → `"25 apples"`), while a hyphen between ordinary words is left alone (`"well-known"`, `"state-of-the-art"`, `"twenty-something"`) ([#1004](https://github.com/drakulavich/kesha-voice-kit/issues/1004)).
 
 ## Text-to-speech
 
-Kesha speaks back in [9 languages](docs/languages.md#text-to-speech). On macOS it picks the voice from the text's own language; on Linux and Windows text detection needs Apple frameworks, so state the language with `--lang <code>` (or the voice with `--voice <id>`) — otherwise the engine default speaks.
+Kesha speaks back in [9 languages](docs/languages.md#text-to-speech). Kokoro runs natively through FluidAudio CoreML/ANE on Apple Silicon and through ONNX on Linux and Windows; Russian uses Vosk-TTS, while `macos-*` system voices need no model download. On macOS Kesha picks the voice from the text's own language; on Linux and Windows, state the language with `--lang <code>` (or the voice with `--voice <id>`) — otherwise the engine default speaks.
 
 ```bash
 kesha install --tts                              # English voices; sizes differ per platform — preview: kesha install --plan

package/docs/product-positioning.md

@@ -21,7 +21,7 @@ The product promise is not "one model for every audio problem." The promise is a
 | Transcribe to structured output for scripts or agents | Stable | `kesha --json audio.ogg`, `kesha --toon audio.ogg`, `kesha --format transcript audio.ogg` |
 | Detect likely audio/text language during transcription | Stable | `--lang`, `--verbose`, JSON/TOON language fields |
 | Skip silence in long audio | Beta | `kesha --vad meeting.m4a`, auto-on for long audio when VAD is installed |
-| Synthesize English or Russian speech locally | Beta | `kesha say "text"`, `kesha install --tts` |
+| Synthesize supported languages locally | Beta | `kesha say "text"`, `kesha install --tts` |
 | Use macOS system voices without model downloads | Stable on macOS | `kesha say --voice macos-*` |
 | Label speakers in meeting transcripts | Preview, darwin-arm64 only | `kesha --json --vad --speakers meeting.m4a` |
 | Integrate with OpenClaw as a local voice model | Stable integration surface, user-configured route | `docs/openclaw.md` |
@@ -35,7 +35,7 @@ The product promise is not "one model for every audio problem." The promise is a
 - Streaming phone-call transcription or real-time conversation infrastructure.
 - Speaker identity across files. Diarization emits per-file cluster IDs, not names or persistent voice profiles.
 - Full SSML coverage. Kesha supports a practical subset for current TTS engines; unsupported tags should fail or warn clearly.
-- Universal TTS across all 25 STT languages. TTS is intentionally scoped to English, Russian, and macOS system voices.
+- Universal TTS across all 25 STT languages. TTS is intentionally scoped to its documented nine-language set and macOS system voices.
 - Model training, fine-tuning, or benchmark leadership on every dataset.
 - Browser/mobile SDKs. The supported product surface is CLI-first, with thin integrations around it.
 
@@ -47,7 +47,7 @@ The product promise is not "one model for every audio problem." The promise is a
 | JSON / TOON / transcript output formats | Stable | Intended for scripts, agents, and downstream tooling. |
 | Audio language detection | Stable | Used for warnings and structured metadata. |
 | VAD preprocessing | Beta | Useful for long or silence-heavy audio; short voice messages stay on the fast path. |
-| English/Russian TTS | Beta | Requires `kesha install --tts`; model cache and first-run cost are expected. |
+| Kokoro and Vosk TTS | Beta | Nine documented languages; requires `kesha install --tts`, so model cache and first-run cost are expected. |
 | macOS AVSpeech voices | Stable on macOS | Zero-install system voices; quality is OS voice quality, not neural TTS. |
 | SSML subset | Preview | `<prosody rate>`, Russian stress markers, and IPA support are engine-specific. |
 | Speaker diarization | Preview | darwin-arm64 only; cluster IDs are stable within one file only. |
@@ -63,8 +63,9 @@ The product promise is not "one model for every audio problem." The promise is a
 | STT | Supported, CoreML path | Not shipped | Supported, ONNX CPU path | Supported, ONNX CPU path | `aarch64-darwin`, `x86_64-linux` |
 | Audio language detection | Supported | Not shipped | Supported | Supported | Supported where the flake builds |
 | VAD | Supported with `kesha install --vad` | Not shipped | Supported with `kesha install --vad` | Supported with `kesha install --vad` | Supported where the engine path includes VAD assets |
-| TTS: English Kokoro | Supported, FluidAudio/CoreML in release builds | Not shipped | Supported, ONNX path | Supported, ONNX path | Supported except where noted in `docs/nix-install.md` |
+| TTS: Kokoro (`en` `es` `fr` `it` `pt`) | Supported, FluidAudio/CoreML in release builds | Not shipped | Supported, ONNX path | Supported, ONNX path | Supported except where noted in `docs/nix-install.md` |
 | TTS: Russian Vosk-TTS | Supported | Not shipped | Supported | Supported | Supported except where noted in `docs/nix-install.md` |
+| TTS: Kokoro (`hi` `ja` `zh`) | Supported, darwin-arm64 only | Not shipped | Not supported | Not supported | `aarch64-darwin` only |
 | macOS system voices | Supported | Not shipped | Not applicable | Not applicable | Supported on `aarch64-darwin` |
 | Speaker diarization | Preview, darwin-arm64 only | Not supported | Not supported | Not supported | Not wired into the Nix build yet |
 | Raycast extension | Supported | Not shipped | Not applicable | Not applicable | Not applicable |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@drakulavich/kesha-voice-kit",
-  "version": "1.28.0",
+  "version": "1.29.1",
   "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": {
@@ -40,11 +40,10 @@
     }
   },
   "keshaEngine": {
-    "version": "1.24.10"
+    "version": "1.24.11"
   },
   "scripts": {
     "test": "bun test --timeout 30000",
-    "conveyor": "bun scripts/backlog.ts",
     "test:watch": "bun test --timeout 30000 --watch",
     "test:unit": "bun test --timeout 30000 tests/unit/",
     "test:cli-fast": "bun test --timeout 30000 tests/unit/ tests/integration/cli-contracts.test.ts tests/integration/error-codes-cli.test.ts tests/integration/say.test.ts",

package/src/engine-targets.ts

@@ -17,17 +17,17 @@ const ENGINE_TARGETS: Record<string, EngineTarget> = {
   "darwin-arm64": {
     assetName: "kesha-engine-darwin-arm64",
     backend: "coreml",
-    sizeBytes: 63_780_320,
+    sizeBytes: 64_514_608,
   },
   "linux-x64": {
     assetName: "kesha-engine-linux-x64",
     backend: "onnx",
-    sizeBytes: 66_101_160,
+    sizeBytes: 66_375_432,
   },
   "win32-x64": {
     assetName: "kesha-engine-windows-x64.exe",
     backend: "onnx",
-    sizeBytes: 65_067_520,
+    sizeBytes: 65_355_264,
   },
 };