@qvac/tts-ggml 0.10.0 → 0.11.0

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@qvac/tts-ggml",
-  "version": "0.10.0",
-  "description": "Text to Speech (TTS) addon for qvac (ggml backend, wrapping the chatterbox + supertonic + parler + cosyvoice3 + audio8 + moss + Pocket engines from tts-cpp)",
+  "version": "0.11.0",
+  "description": "Text to Speech (TTS) addon for qvac (ggml backend, wrapping the chatterbox + supertonic + parler + cosyvoice3 + audio8 + moss + moss-sfx + moss-speech + Pocket engines from tts-cpp)",
   "addon": true,
   "engines": {
     "bare": ">=1.19.0"
@@ -53,8 +53,8 @@
   },
   "scripts": {
     "build": "npm run build:ts && npm run build:native",
-    "build:native": "bare-make generate && bare-make build && bare-make install",
-    "build:native:cuda": "bare-make generate -D ENABLE_CUDA=ON && bare-make build && bare-make install",
+    "build:native": "bare-make generate -D BUILD_FUZZING=OFF && bare-make build && bare-make install",
+    "build:native:cuda": "bare-make generate -D ENABLE_CUDA=ON -D BUILD_FUZZING=OFF && bare-make build && bare-make install",
     "build:cuda": "npm run build:ts && npm run build:native:cuda",
     "build:ts": "tsc -p tsconfig.build.json",
     "build:pack": "npm run build:ts && mkdir -p dist && npm pack --pack-destination dist",
@@ -80,8 +80,10 @@
     "example:parler-enhanced": "bare examples/parler-enhanced.js \"Hello from enhanced Parler.\"",
     "example:audio8": "bare examples/audio8-tts.js \"Audio eight speaks on-device.\"",
     "example:moss": "bare examples/moss-tts.js \"MOSS speaks on-device.\"",
+    "example:moss-sfx": "bare examples/moss-sound-effect.js \"Heavy rain falling on a tin roof with distant thunder.\" 5",
+    "example:moss-speech": "bare examples/moss-speech.js test/reference-audio/jfk.wav",
     "lint": "npm run lint:js && npm run lint:ts && npm run test:types",
-    "lint:js": "prettier --check \"test/**/*.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"binding.js\" \"addon-unavailable.js\" && lunte \"test/**/*.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"binding.js\" \"addon-unavailable.js\"",
+    "lint:js": "prettier --check \"test/**/*.js\" \"scripts/run-cpp-fuzz.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"scripts/__tests__/run-cpp-fuzz.test.js\" \"binding.js\" \"addon-unavailable.js\" && lunte \"test/**/*.js\" \"scripts/run-cpp-fuzz.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"scripts/__tests__/run-cpp-fuzz.test.js\" \"binding.js\" \"addon-unavailable.js\"",
     "lint:ts": "eslint \"src/**/*.ts\" --max-warnings=0",
     "lint:ts:fix": "eslint \"src/**/*.ts\" --fix",
     "format": "prettier --write \"test/**/*.js\" \"scripts/__tests__/package-contract.test.js\" \"scripts/__tests__/binding-loader.test.js\" \"binding.js\" \"addon-unavailable.js\"",
@@ -89,15 +91,28 @@
     "test:node": "node --test scripts/__tests__/*.test.js",
     "test": "npm run test:types && npm run test:unit && npm run test:integration",
     "test:integration": "npm run build:ts && brittle-bare test/integration/**/*.test.js",
+    "test:integration:smoke": "npm run build:ts && brittle-bare test/integration/supertonic.test.js",
     "test:benchmark:rtf": "npm run build:ts && brittle-bare test/benchmark/rtf-benchmark.test.js",
     "test:benchmark:rtf:matrix": "node scripts/run-rtf-benchmark-matrix.js",
     "test:benchmark:streaming": "npm run build:ts && brittle-bare test/benchmark/streaming-benchmark.test.js",
     "test:mobile:generate": "node scripts/generate-mobile-integration-tests.js",
     "test:mobile:validate": "node scripts/validate-mobile-tests.js",
-    "test:cpp:build": "bare-make generate -D BUILD_TESTING=ON && bare-make build --target tts_ggml_tests && bare-make install",
+    "test:cpp:build": "bare-make generate -D BUILD_TESTING=ON -D BUILD_FUZZING=OFF && bare-make build --target tts_ggml_tests && bare-make install",
     "test:cpp:run": "cd build && LLVM_PROFILE_FILE=default.profraw ./tts_ggml_tests --gtest_output=xml:cpp-test-results.xml",
     "test:cpp": "npm run test:cpp:build && npm run test:cpp:run",
-    "coverage:cpp:build": "bare-make generate -D BUILD_TESTING=ON -D ENABLE_COVERAGE=ON && bare-make build --target tts_ggml_tests",
+    "fuzz:build": "bare-make generate -D BUILD_FUZZING=ON -D BUILD_TESTING=OFF -D FUZZTEST_FUZZING_MODE=OFF && bare-make build --target tts-config-parse-fuzz",
+    "fuzz:build:continuous": "bare-make generate -D BUILD_FUZZING=ON -D BUILD_TESTING=OFF -D FUZZTEST_FUZZING_MODE=ON && bare-make build --target tts-config-parse-fuzz",
+    "fuzz:run": "node scripts/run-cpp-fuzz.js",
+    "fuzz": "npm run fuzz:build && npm run fuzz:run",
+    "fuzz:continuous": "npm run fuzz:build:continuous && node scripts/run-cpp-fuzz.js --continuous",
+    "test:cpp:fuzz:configure": "bare-make generate -D BUILD_TESTING=ON -D BUILD_FUZZING=ON -D FUZZTEST_FUZZING_MODE=OFF",
+    "test:cpp:fuzz:build:tests": "bare-make build --target tts_ggml_tests",
+    "test:cpp:fuzz:build:fuzz": "bare-make build --target tts-config-parse-fuzz",
+    "test:cpp:fuzz:build": "npm run test:cpp:fuzz:configure && npm run test:cpp:fuzz:build:tests && npm run test:cpp:fuzz:build:fuzz",
+    "test:cpp:fuzz:run": "npm run test:scripts && npm run test:cpp:run && node scripts/run-cpp-fuzz.js",
+    "test:cpp:fuzz": "npm run test:cpp:fuzz:build && npm run test:cpp:fuzz:run",
+    "test:scripts": "node --test scripts/__tests__/run-cpp-fuzz.test.js",
+    "coverage:cpp:build": "bare-make generate -D BUILD_TESTING=ON -D ENABLE_COVERAGE=ON -D BUILD_FUZZING=OFF && bare-make build --target tts_ggml_tests",
     "coverage:cpp:summary": "cd build && llvm-cov report ./tts_ggml_tests --instr-profile=coverage.profdata -ignore-filename-regex='(tests|build|node_modules|gtest|gmock|\\.vcpkg|/usr)/' > coverage-summary.txt",
     "coverage:cpp:report": "cd build && ls -lha && llvm-profdata merge -sparse default.profraw -o coverage.profdata && llvm-cov show ./tts_ggml_tests -instr-profile=coverage.profdata -format=html -output-dir=coverage-html -ignore-filename-regex='(tests|build|node_modules|gtest|gmock|\\.vcpkg|/usr)/' && llvm-cov export ./tts_ggml_tests -instr-profile=coverage.profdata -format=lcov -ignore-filename-regex='(tests|build|node_modules|gtest|gmock|\\.vcpkg|/usr)/' > lcov.info && npm run coverage:cpp:summary",
     "coverage:cpp": "npm run coverage:cpp:build && npm run test:cpp:run && npm run coverage:cpp:report",
@@ -107,6 +122,7 @@
     "convert-models": "bash scripts/convert-models.sh",
     "setup-models": "bash scripts/setup-venv.sh && bash scripts/convert-models.sh",
     "download-models:registry": "node scripts/download-tts-ggml-models.js",
+    "download-models:smoke": "node scripts/download-tts-ggml-models.js --group supertonic",
     "example:pocket": "bare examples/pocket-tts.js",
     "test:pocket": "npm run build:ts && brittle-bare test/integration/pocket.integration.test.js"
   },
@@ -125,6 +141,7 @@
     "scripts/download-tts-ggml-models.js",
     "test/integration",
     "test/mobile",
+    "test/pocket-worklet",
     "test/reference-audio",
     "test/utils/runTTS.js",
     "test/utils/runChatterboxTTS.js",
@@ -195,7 +212,9 @@
     "cmake-npm": "^1.1.0",
     "bare-url": "^2.1.6",
     "bare-https": "^3.0.0",
-    "bare-stream": "^2.7.0"
+    "bare-stream": "^2.7.0",
+    "bare-subprocess": "^6.2.1",
+    "brittle": "^3.17.0"
   },
   "dependencies": {
     "@qvac/error": "^0.1.0",
@@ -208,16 +227,14 @@
     "bare-path": "^3.0.0",
     "bare-process": "^4.2.2",
     "bare-stream": "^2.7.0",
-    "bare-subprocess": "^5.2.1",
-    "bare-url": "^2.4.3",
-    "brittle": "^3.17.0"
+    "bare-url": "^2.4.3"
   },
   "types": "index.d.ts",
   "optionalDependencies": {
-    "@qvac/tts-ggml-linux-x64": "0.10.0",
-    "@qvac/tts-ggml-linux-arm64": "0.10.0",
-    "@qvac/tts-ggml-darwin-arm64": "0.10.0",
-    "@qvac/tts-ggml-darwin-x64": "0.10.0",
-    "@qvac/tts-ggml-win32-x64": "0.10.0"
+    "@qvac/tts-ggml-linux-x64": "0.11.0",
+    "@qvac/tts-ggml-linux-arm64": "0.11.0",
+    "@qvac/tts-ggml-darwin-arm64": "0.11.0",
+    "@qvac/tts-ggml-darwin-x64": "0.11.0",
+    "@qvac/tts-ggml-win32-x64": "0.11.0"
   }
 }

package/test/integration/addon.test.js

@@ -16,11 +16,11 @@ const { ensureChatterboxModels, ensureWhisperModel } = require('../utils/downloa
 const { loadWhisper, runWhisper } = require('../utils/runWhisper')
 const { recordTtsStats } = require('../utils/perf-helper')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { NO_GPU, GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 const platform = os.platform()
 const isMobile = platform === 'ios' || platform === 'android'
 const isDarwin = platform === 'darwin'
-const forceNoGpu = os.getEnv('NO_GPU') === 'true'
 
 const INPUT_SENTENCES = (isMobile ? 'short' : os.getEnv('INPUT_SENTENCES')) || 'short'
 const useSplit = INPUT_SENTENCES !== 'short'
@@ -42,7 +42,7 @@ function getEnglishSentences() {
 
 test(
   'Chatterbox TTS (ggml): English synthesis + optional WER verification',
-  { timeout: 1800000 },
+  { timeout: 1800000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const modelsDir = path.join(baseDir, 'models')
@@ -86,7 +86,8 @@ test(
     const model = await loadChatterboxTTS({
       threads: TTS_TEST_THREADS,
       modelDir: resolvedModelDir,
-      language: 'en'
+      language: 'en',
+      useGPU: GPU_ONLY_USE_GPU
     })
     t.ok(model, 'Chatterbox (ggml) model should be loaded')
 
@@ -254,7 +255,7 @@ test(
     const model = new TTSGgml({
       threads: TTS_TEST_THREADS,
       files: { modelDir: download.targetDir },
-      config: { language: 'en', ...(forceNoGpu ? { useGPU: false } : {}) },
+      config: { language: 'en', ...(NO_GPU ? { useGPU: false } : {}) },
       opts: { stats: true }
     })
 
@@ -315,7 +316,7 @@ test(
       threads: TTS_TEST_THREADS,
       files: { modelDir: download.targetDir },
       referenceAudio: resolveRefWavPath({}),
-      config: { language: 'en', outputSampleRate: 16000, ...(forceNoGpu ? { useGPU: false } : {}) },
+      config: { language: 'en', outputSampleRate: 16000, ...(NO_GPU ? { useGPU: false } : {}) },
       opts: { stats: true }
     })
     await model.load()
@@ -344,7 +345,7 @@ test(
 
 test(
   'Chatterbox TTS (ggml): native C++ chunk streaming via streamChunkTokens',
-  { timeout: 600000 },
+  { timeout: 600000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const modelsDir = path.join(baseDir, 'models')
@@ -371,7 +372,7 @@ test(
       streamChunkTokens: 25,
       streamFirstChunkTokens: 10,
       cfmSteps: 1,
-      config: { language: 'en', ...(forceNoGpu ? { useGPU: false } : {}) },
+      config: { language: 'en', useGPU: GPU_ONLY_USE_GPU },
       opts: { stats: true }
     })
     await model.load()

package/test/integration/chatterbox-mtl.test.js

@@ -26,13 +26,15 @@ const {
 } = require('../utils/downloadModel')
 const { recordTtsStats } = require('../utils/perf-helper')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { NO_GPU, GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 const platform = os.platform()
 const isMobile = platform === 'ios' || platform === 'android'
 
-// Language coverage test, not a GPU policy test: rely on the package
-// default (`useGPU: false`) rather than opting into GPU here.
-// Tests that *are* about GPU live in gpu-smoke.test.js.
+// Language coverage, not a GPU policy test. A single MTL synthesis already
+// takes close to a minute on the CPU-only runners, so these tests run on the
+// GPU only and are skipped there (NO_GPU=true). Tests that *are* about GPU live
+// in gpu-smoke.test.js.
 
 function getBaseDir() {
   return isMobile && global.testDir ? global.testDir : '.'
@@ -83,7 +85,7 @@ async function loadChatterboxMtlTTS(params) {
 
 test(
   'Chatterbox MTL TTS (ggml): synthesizes across es/fr/de with shared engine',
-  { timeout: 1800000 },
+  { timeout: 1800000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const download = await ensureChatterboxMtlModels({ targetDir: path.join(baseDir, 'models') })
@@ -98,7 +100,8 @@ test(
       modelDir: download.targetDir,
       t3ModelPath: path.join(download.targetDir, 'chatterbox-t3-mtl.gguf'),
       s3genModelPath: path.join(download.targetDir, 'chatterbox-s3gen-mtl.gguf'),
-      language: MTL_SENTENCES[0].lang
+      language: MTL_SENTENCES[0].lang,
+      useGPU: GPU_ONLY_USE_GPU
     })
     try {
       for (let i = 0; i < MTL_SENTENCES.length; i++) {
@@ -166,7 +169,7 @@ test(
 
 test(
   'Chatterbox MTL TTS (ggml): synthesizes Japanese with MeCab dictionary',
-  { timeout: 1800000 },
+  { timeout: 1800000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const download = await ensureChatterboxMtlModels({ targetDir: path.join(baseDir, 'models') })
@@ -188,7 +191,8 @@ test(
       t3ModelPath: path.join(download.targetDir, 'chatterbox-t3-mtl.gguf'),
       s3genModelPath: path.join(download.targetDir, 'chatterbox-s3gen-mtl.gguf'),
       mecabDictDir: mecab.dir,
-      language: 'ja'
+      language: 'ja',
+      useGPU: GPU_ONLY_USE_GPU
     })
     try {
       const t0 = Date.now()
@@ -233,7 +237,7 @@ test(
 
 test(
   'Chatterbox MTL TTS (ggml): synthesizes Chinese with Cangjie table',
-  { timeout: 1800000 },
+  { timeout: 1800000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const download = await ensureChatterboxMtlModels({ targetDir: path.join(baseDir, 'models') })
@@ -255,7 +259,8 @@ test(
       t3ModelPath: path.join(download.targetDir, 'chatterbox-t3-mtl.gguf'),
       s3genModelPath: path.join(download.targetDir, 'chatterbox-s3gen-mtl.gguf'),
       cangjieTsvPath: cangjie.path,
-      language: 'zh'
+      language: 'zh',
+      useGPU: GPU_ONLY_USE_GPU
     })
     try {
       const t0 = Date.now()

package/test/integration/chatterbox-speed.test.js

@@ -12,6 +12,7 @@ const path = require('bare-path')
 const { loadChatterboxTTS, runChatterboxTTS } = require('../utils/runChatterboxTTS')
 const { ensureChatterboxModels } = require('../utils/downloadModel')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { NO_GPU, GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 function getBaseDir() {
   const platform = os.platform()
@@ -30,7 +31,7 @@ const EXPECTATION = {
 
 test(
   'Chatterbox TTS (ggml): speed config adjusts duration, preserves backward compat',
-  { timeout: 1800000 },
+  { timeout: 1800000, skip: NO_GPU },
   async (t) => {
     const modelsDir = path.join(getBaseDir(), 'models')
 
@@ -52,7 +53,8 @@ test(
         threads: TTS_TEST_THREADS,
         modelDir,
         language: 'en',
-        speed
+        speed,
+        useGPU: GPU_ONLY_USE_GPU
       })
       try {
         const r = await runChatterboxTTS(model, { text: TEXT }, EXPECTATION)

package/test/integration/cosyvoice3-clone.test.js

@@ -8,7 +8,6 @@
 const fs = require('bare-fs')
 const os = require('bare-os')
 const path = require('bare-path')
-const proc = require('bare-process')
 const test = require('brittle')
 const TTSGgml = require('@qvac/tts-ggml')
 
@@ -20,10 +19,11 @@ const {
 } = require('../utils/downloadModel')
 const { resolveRefWavPath } = require('../utils/runChatterboxTTS')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { NO_GPU, GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 const platform = os.platform()
 const isMobile = platform === 'ios' || platform === 'android'
-const useGPU = !isMobile && proc.env.NO_GPU === 'false'
+const useGPU = GPU_ONLY_USE_GPU
 
 function getBaseDir() {
   return isMobile && global.testDir ? global.testDir : '.'
@@ -99,7 +99,7 @@ async function synthOnce(t, loadParams, label) {
 
 test(
   'CosyVoice3 cloning (ggml): baked, zero-shot and cross-lingual all differ at one seed',
-  { timeout: 1800000 },
+  { timeout: 1800000, skip: NO_GPU },
   async (t) => {
     const modelDir = await ensureCloneReadyModelDir(t)
     if (!modelDir) return

package/test/integration/cosyvoice3-lavasr.test.js

@@ -7,7 +7,9 @@
 // CosyVoice3's ~2.4 GB model dir, while this one is listed on the dedicated
 // 'cosyvoice' row that pre-stages it (see test/mobile/test-groups.json).
 //
-// Text is kept short to bound CPU LM-decode time in CI, as in cosyvoice3.test.js.
+// As in cosyvoice3.test.js, the model-backed tests run on the GPU only and are
+// skipped on CPU-only runners (NO_GPU=true): even one short CosyVoice3 CPU
+// synthesis costs minutes there.
 //
 // Stage the enhancer GGUF via scripts/convert-lavasr-enhancer-to-gguf.py (from
 // the public LavaSRcpp ONNX release) into models/lavasr/lavasr-enhancer.gguf,
@@ -25,6 +27,7 @@ const {
   ensureCosyvoiceModel
 } = require('../utils/downloadModel')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { NO_GPU, GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 const platform = os.platform()
 const isMobile = platform === 'ios' || platform === 'android'
@@ -118,7 +121,7 @@ async function stageCosyvoice(t, baseDir) {
 
 test(
   'CosyVoice3 + LavaSR enhancer (batch) reports 48 kHz enhanced output',
-  { timeout: 900000, skip: SKIP_COSYVOICE },
+  { timeout: 900000, skip: SKIP_COSYVOICE || NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const enh = await stageEnhancer(t, baseDir)
@@ -130,7 +133,7 @@ test(
       threads: TTS_TEST_THREADS,
       engine: TTSGgml.ENGINE_COSYVOICE3,
       files: { cosyvoiceModelDir: dl.modelDir, lavasrEnhancer: enh.path },
-      config: { language: 'en', useGPU: false },
+      config: { language: 'en', useGPU: GPU_ONLY_USE_GPU },
       opts: { stats: true }
     })
     await model.load()
@@ -141,8 +144,8 @@ test(
       t.ok(r.stats, 'runtimeStats returned (constructed with stats:true)')
       t.is(
         r.stats.enhancerBackendDevice,
-        0,
-        'enhancer loaded and ran on CPU (enhancerBackendDevice=0, not -1 = never loaded)'
+        r.stats.backendDevice,
+        'enhancer loaded and ran on the engine device (not -1 = never loaded)'
       )
     } finally {
       try {
@@ -154,7 +157,7 @@ test(
 
 test(
   'CosyVoice3 + LavaSR enhancer + native chunk streaming emits 48 kHz chunks',
-  { timeout: 900000, skip: SKIP_COSYVOICE },
+  { timeout: 900000, skip: SKIP_COSYVOICE || NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const enh = await stageEnhancer(t, baseDir)
@@ -167,7 +170,7 @@ test(
       engine: TTSGgml.ENGINE_COSYVOICE3,
       files: { cosyvoiceModelDir: dl.modelDir, lavasrEnhancer: enh.path },
       streamChunkTokens: 25,
-      config: { language: 'en', useGPU: false },
+      config: { language: 'en', useGPU: GPU_ONLY_USE_GPU },
       opts: { stats: true }
     })
     await model.load()
@@ -191,7 +194,7 @@ test(
 
 test(
   'CosyVoice3 + enhancer + streaming resamples every chunk to outputSampleRate',
-  { timeout: 900000, skip: SKIP_COSYVOICE },
+  { timeout: 900000, skip: SKIP_COSYVOICE || NO_GPU },
   async (t) => {
     // A non-native rate while streaming is only legal because the enhancer's
     // overlap-reprocess window folds the resample in seam-free. Without this the
@@ -208,7 +211,7 @@ test(
       engine: TTSGgml.ENGINE_COSYVOICE3,
       files: { cosyvoiceModelDir: dl.modelDir, lavasrEnhancer: enh.path },
       streamChunkTokens: 25,
-      config: { language: 'en', useGPU: false, outputSampleRate: 16000 },
+      config: { language: 'en', useGPU: GPU_ONLY_USE_GPU, outputSampleRate: 16000 },
       opts: { stats: true }
     })
     await model.load()
@@ -228,7 +231,7 @@ test(
 
 test(
   'CosyVoice3 + LavaSR denoiser changes the emitted PCM',
-  { timeout: 900000, skip: SKIP_COSYVOICE },
+  { timeout: 900000, skip: SKIP_COSYVOICE || NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const dl = await stageCosyvoice(t, baseDir)
@@ -247,7 +250,7 @@ test(
         threads: TTS_TEST_THREADS,
         engine: TTSGgml.ENGINE_COSYVOICE3,
         files: Object.assign({ cosyvoiceModelDir: dl.modelDir }, files),
-        config: { language: 'en', useGPU: false, seed: SEED },
+        config: { language: 'en', useGPU: GPU_ONLY_USE_GPU, seed: SEED },
         opts: { stats: true }
       })
       await model.load()
@@ -294,6 +297,7 @@ test(
     if (!dl) return
 
     const badEnhancer = path.join(baseDir, 'models', 'lavasr', 'invalid-enhancer.gguf')
+    fs.mkdirSync(path.dirname(badEnhancer), { recursive: true })
     fs.writeFileSync(badEnhancer, 'not a gguf')
 
     const model = new TTSGgml({

package/test/integration/cosyvoice3.test.js

@@ -3,8 +3,10 @@
 // CosyVoice3 engine integration smoke: real synthesis in the desktop
 // `run-integration-tests` lane.  Mirrors parler.test.js (download model → build
 // TTSGgml → run → assert) but for the directory-consuming CosyVoice3 engine
-// (Qwen2 speech LM + DiT flow + CausalHiFT vocoder; native 24 kHz, CPU-only).
-// Text is kept SHORT to bound CPU LM-decode time in CI.
+// (Qwen2 speech LM + DiT flow + CausalHiFT vocoder; native 24 kHz).
+// Even one short CPU synthesis costs minutes on the CPU-only runners, so these
+// tests run on the GPU only and are skipped there (NO_GPU=true); the CPU crash
+// check is the CosyVoice3 CPU smoke in gpu-smoke.test.js.
 
 const os = require('bare-os')
 const path = require('bare-path')
@@ -13,6 +15,7 @@ const test = require('brittle')
 const { loadCosyvoiceTTS, runCosyvoiceTTS } = require('../utils/runCosyvoiceTTS')
 const { ensureCosyvoiceModel } = require('../utils/downloadModel')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { NO_GPU, GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 const platform = os.platform()
 const isMobile = platform === 'ios' || platform === 'android'
@@ -27,7 +30,7 @@ const MODEL_MISSING =
 
 test(
   'CosyVoice3 TTS (ggml): outputSampleRate=16000 resamples and reports 16 kHz',
-  { timeout: 600000 },
+  { timeout: 600000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const download = await ensureCosyvoiceModel({
@@ -43,7 +46,8 @@ test(
     const model = await loadCosyvoiceTTS({
       threads: TTS_TEST_THREADS,
       cosyvoiceModelDir: download.modelDir,
-      outputSampleRate: 16000
+      outputSampleRate: 16000,
+      useGPU: GPU_ONLY_USE_GPU
     })
     try {
       const text = 'Hello from CosyVoice.'
@@ -70,7 +74,7 @@ function samplesEqual(a, b) {
 
 test(
   'CosyVoice3 TTS (ggml): emotion changes the audio; per-call switch needs no reload',
-  { timeout: 900000 },
+  { timeout: 900000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const download = await ensureCosyvoiceModel({
@@ -87,7 +91,8 @@ test(
     const model = await loadCosyvoiceTTS({
       threads: TTS_TEST_THREADS,
       cosyvoiceModelDir: download.modelDir,
-      seed: 42
+      seed: 42,
+      useGPU: GPU_ONLY_USE_GPU
     })
     try {
       const text = 'Hello from CosyVoice.'
@@ -172,7 +177,7 @@ test(
 
 test(
   'CosyVoice3 TTS (ggml): instruct conditioning produces audio',
-  { timeout: 600000 },
+  { timeout: 600000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const download = await ensureCosyvoiceModel({
@@ -186,7 +191,8 @@ test(
     const model = await loadCosyvoiceTTS({
       threads: TTS_TEST_THREADS,
       cosyvoiceModelDir: download.modelDir,
-      emotion: 'happy'
+      emotion: 'happy',
+      useGPU: GPU_ONLY_USE_GPU
     })
     try {
       const text = 'Hello from CosyVoice.'

package/test/integration/fit.test.js

@@ -141,14 +141,18 @@ test('a LavaSR stage is sized beside the projection', { timeout: TEST_TIMEOUT_MS
   t.ok(fit.lavasrFileBytes < fit.deviceBytes, 'reported beside the projection')
 })
 
-test('a voice with no fitter is an outcome, not a throw', (t) => {
+test('MOSS routes to its fitter when a model cannot be read', (t) => {
   const fit = TTSGgml.assessFit({
     engineType: 'moss',
-    mossBackbonePath: '/models/moss-tts-delay.gguf'
+    mossBackbonePath: '/models/moss-tts-delay.gguf',
+    mossCodecDecoderPath: '/models/moss-codec-decoder.gguf',
+    promptRows: 128,
+    referenceSamples: 0,
+    streaming: false
   })
 
   t.is(fit.status, 'error')
-  t.is(fit.reason, 'unsupported-engine')
+  t.is(fit.reason, 'model-unreadable')
   t.is(fit.modelVariant, 'moss')
 })
 

package/test/integration/lavasr-enhancer.test.js

@@ -33,6 +33,7 @@ const {
 } = require('../utils/downloadModel')
 const { resolveRefWavPath } = require('../utils/runChatterboxTTS')
 const { TTS_TEST_THREADS } = require('../utils/testThreads')
+const { GPU_ONLY_USE_GPU } = require('../utils/gpuOnly')
 
 const platform = os.platform()
 const isMobile = platform === 'ios' || platform === 'android'
@@ -632,7 +633,7 @@ test(
 
 test(
   'Chatterbox + LavaSR enhancer + native chunk streaming emits 48 kHz chunks',
-  { timeout: 900000 },
+  { timeout: 900000, skip: NO_GPU },
   async (t) => {
     const baseDir = getBaseDir()
     const enh = await ensureLavaSREnhancerGguf({
@@ -662,7 +663,7 @@ test(
       },
       referenceAudio: resolveRefWavPath({}),
       streamChunkTokens: 25, // native chunk streaming + enhancer (the path)
-      config: { language: 'en', useGPU: false },
+      config: { language: 'en', useGPU: GPU_ONLY_USE_GPU },
       opts: { stats: true }
     })
     await model.load()

package/test/integration/moss-fit.test.js

@@ -0,0 +1,57 @@
+'use strict'
+
+const path = require('bare-path')
+const process = require('bare-process')
+const test = require('brittle')
+const TTSGgml = require('@qvac/tts-ggml')
+
+test('MOSS fit requires explicit, representable workload counts', (t) => {
+  const request = {
+    engineType: 'moss',
+    mossBackbonePath: '/models/moss-tts-delay.gguf',
+    mossCodecDecoderPath: '/models/moss-codec-decoder.gguf',
+    promptRows: 128,
+    referenceSamples: 0,
+    streaming: false
+  }
+  for (const invalid of [NaN, Infinity, -1, 0.5, 2147483648]) {
+    t.exception(() => TTSGgml.assessFit({ ...request, promptRows: invalid }))
+    t.exception(() => TTSGgml.assessFit({ ...request, referenceSamples: invalid }))
+  }
+  t.exception(() => TTSGgml.assessFit({ ...request, promptRows: undefined }))
+  t.exception(() => TTSGgml.assessFit({ ...request, streaming: undefined }))
+  t.exception(() => TTSGgml.assessFit({ ...request, threads: -1 }))
+  t.exception(() => TTSGgml.assessFit({ ...request, durationTokens: -1 }))
+  t.exception(() => TTSGgml.assessFit({ ...request, streamChunkTokens: -1 }))
+  t.exception(() => TTSGgml.assessFit({ ...request, useGPU: true, nGpuLayers: 0 }))
+})
+
+test('MOSS metadata-only fit covers batch, streaming and cloning', (t) => {
+  const modelDir = process.env.QVAC_TEST_MOSS_MODEL_DIR
+  if (!modelDir) {
+    t.comment('Set QVAC_TEST_MOSS_MODEL_DIR to exercise real MOSS GGUFs')
+    return
+  }
+  const request = {
+    engineType: 'moss',
+    mossBackbonePath: path.join(modelDir, 'moss-tts-delay-f16.gguf'),
+    mossCodecDecoderPath: path.join(modelDir, 'moss-codec-decoder-f16.gguf'),
+    promptRows: 128,
+    referenceSamples: 0,
+    streaming: false
+  }
+  const batch = TTSGgml.assessFit(request)
+  t.not(batch.status, 'error', batch.report)
+  t.ok(batch.weightsBytes > 0)
+  t.ok(batch.lmComputeBytes > 0)
+  t.ok(batch.codecComputeBytes > 0)
+  const stream = TTSGgml.assessFit({ ...request, streaming: true })
+  t.not(stream.status, 'error', stream.report)
+  const cloning = TTSGgml.assessFit({
+    ...request,
+    mossCodecEncoderPath: path.join(modelDir, 'moss-codec-encoder-f16.gguf'),
+    referenceSamples: 24000
+  })
+  t.not(cloning.status, 'error', cloning.report)
+  t.ok(cloning.weightsBytes > batch.weightsBytes)
+})