Fix/whisper silence hallucination

.github/workflows/integration-mobile-test-qvac-lib-infer-whispercpp.yml

@@ -931,7 +931,7 @@ jobs:
             BUNDLE_ID="${{ env.APP_BUNDLE_ID }}"
             # Android wdio config with crash detection (bail:0 = continue on test failures, crash = process.exit)
             # Timeout set to 15 minutes (900000ms) for audio transcription tests (whisper models can be slow)
-            WDIO_CONFIG='exports.config={runner:"local",hostname:"127.0.0.1",port:4723,path:"/wd/hub",specs:["*.spec.js","*.test.js"],maxInstances:1,bail:0,capabilities:[{platformName:"Android","appium:automationName":"UiAutomator2","appium:appPackage":"'${{ env.APP_BUNDLE_ID }}'","appium:appActivity":"'${{ env.APP_BUNDLE_ID }}'.MainActivity","appium:newCommandTimeout":300,"appium:autoGrantPermissions":true,"appium:autoAcceptAlerts":true,"appium:noReset":true,"appium:dontStopAppOnReset":true,"appium:forceAppLaunch":false}],logLevel:"debug",waitforTimeout:120000,connectionRetryTimeout:30000,connectionRetryCount:3,services:[],framework:"mocha",reporters:["spec"],mochaOpts:{ui:"bdd",timeout:900000},before:async function(capabilities,specs,browser){const BUNDLE_ID="'${{ env.APP_BUNDLE_ID }}'";global.appCrashed=false;global.checkAppCrash=async(stage)=>{try{const state=await browser.queryAppState(BUNDLE_ID);console.log("["+stage+"] App state: "+state+" (4=foreground,3=background,1=not running)");if(state<3){console.error("\\n🛑 APP CRASHED at "+stage+"! State="+state);console.error("Check device logs for BareKit/native errors.\\n");global.appCrashed=true;process.exit(1);}return state;}catch(e){console.log("["+stage+"] queryAppState error: "+e.message);return-1;}};console.log("Checking initial app state...");await global.checkAppCrash("startup");console.log("Waiting for app to initialize...");await browser.pause(5000);await global.checkAppCrash("after-pause");const initText=await browser.$("android=new UiSelector().textContains(\\"INITIALIZED\\")");await initText.waitForDisplayed({timeout:60000});await global.checkAppCrash("after-init");console.log("App initialized, clicking Run Automated Tests...");const button=await browser.$("android=new UiSelector().textContains(\\"Run Automated Tests\\")");await button.waitForDisplayed({timeout:15000});await button.click();console.log("Button clicked!");await browser.pause(5000);await global.checkAppCrash("after-click");},afterTest:async function(test,context,{error}){if(global.appCrashed)return;await global.checkAppCrash("after-test:"+test.title);}};'
+            WDIO_CONFIG='exports.config={runner:"local",hostname:"127.0.0.1",port:4723,path:"/wd/hub",specs:["tests/*.spec.js","tests/*.test.js"],maxInstances:1,bail:0,capabilities:[{platformName:"Android","appium:automationName":"UiAutomator2","appium:appPackage":"'${{ env.APP_BUNDLE_ID }}'","appium:appActivity":"'${{ env.APP_BUNDLE_ID }}'.MainActivity","appium:newCommandTimeout":300,"appium:autoGrantPermissions":true,"appium:autoAcceptAlerts":true,"appium:noReset":true,"appium:dontStopAppOnReset":true,"appium:forceAppLaunch":false}],logLevel:"debug",waitforTimeout:120000,connectionRetryTimeout:30000,connectionRetryCount:3,services:[],framework:"mocha",reporters:["spec"],mochaOpts:{ui:"bdd",timeout:900000},before:async function(capabilities,specs,browser){const BUNDLE_ID="'${{ env.APP_BUNDLE_ID }}'";global.appCrashed=false;global.checkAppCrash=async(stage)=>{try{const state=await browser.queryAppState(BUNDLE_ID);console.log("["+stage+"] App state: "+state+" (4=foreground,3=background,1=not running)");if(state<3){console.error("\\n🛑 APP CRASHED at "+stage+"! State="+state);console.error("Check device logs for BareKit/native errors.\\n");global.appCrashed=true;process.exit(1);}return state;}catch(e){console.log("["+stage+"] queryAppState error: "+e.message);return-1;}};console.log("Checking initial app state...");await global.checkAppCrash("startup");console.log("Waiting for app to initialize...");await browser.pause(5000);await global.checkAppCrash("after-pause");const initText=await browser.$("android=new UiSelector().textContains(\\"INITIALIZED\\")");await initText.waitForDisplayed({timeout:60000});await global.checkAppCrash("after-init");console.log("App initialized, clicking Run Automated Tests...");const button=await browser.$("android=new UiSelector().textContains(\\"Run Automated Tests\\")");await button.waitForDisplayed({timeout:15000});await button.click();console.log("Button clicked!");await browser.pause(5000);await global.checkAppCrash("after-click");},afterTest:async function(test,context,{error}){if(global.appCrashed)return;await global.checkAppCrash("after-test:"+test.title);}};'
           else
             PLATFORM="iOS"
             AUTOMATION="XCUITest"
@@ -941,7 +941,7 @@ jobs:
             # iOS wdio config with crash detection (bail:0 = continue on test failures, crash = process.exit)
             # usePrebuiltWDA uses Device Farm's pre-built WebDriverAgent
             # Timeout set to 15 minutes (900000ms) for audio transcription tests (whisper models can be slow)
-            WDIO_CONFIG='exports.config={runner:"local",hostname:"127.0.0.1",port:4723,path:"/wd/hub",specs:["*.spec.js","*.test.js"],maxInstances:1,bail:0,capabilities:[{platformName:"iOS","appium:automationName":"XCUITest","appium:bundleId":"'${{ env.APP_BUNDLE_ID }}'","appium:newCommandTimeout":300,"appium:noReset":true,"appium:forceAppLaunch":false,"appium:usePrebuiltWDA":true,"appium:wdaLocalPort":8100,"appium:showIOSLog":true,"appium:realDeviceLogger":"/usr/local/lib/node_modules/appium/node_modules/deviceconsole/deviceconsole"}],logLevel:"debug",waitforTimeout:120000,connectionRetryTimeout:30000,connectionRetryCount:3,services:[],framework:"mocha",reporters:["spec"],mochaOpts:{ui:"bdd",timeout:900000},before:async function(capabilities,specs,browser){const BUNDLE_ID="'${{ env.APP_BUNDLE_ID }}'";global.appCrashed=false;global.checkAppCrash=async(stage)=>{try{const state=await browser.queryAppState(BUNDLE_ID);console.log("["+stage+"] App state: "+state+" (4=foreground,3=background,1=not running)");if(state<3){console.error("\\n🛑 APP CRASHED at "+stage+"! State="+state);console.error("Check device logs for BareKit/native errors.\\n");global.appCrashed=true;process.exit(1);}return state;}catch(e){console.log("["+stage+"] queryAppState error: "+e.message);return-1;}};console.log("Checking initial app state...");await global.checkAppCrash("startup");console.log("Waiting for app to initialize...");await browser.pause(5000);await global.checkAppCrash("after-pause");const initText=await browser.$("-ios predicate string:label CONTAINS \\"INITIALIZED\\"");await initText.waitForDisplayed({timeout:60000});await global.checkAppCrash("after-init");console.log("App initialized, clicking Run Automated Tests...");const button=await browser.$("-ios predicate string:label CONTAINS \\"Run Automated Tests\\"");await button.waitForDisplayed({timeout:15000});await button.click();console.log("Button clicked!");await browser.pause(5000);await global.checkAppCrash("after-click");},afterTest:async function(test,context,{error}){if(global.appCrashed)return;await global.checkAppCrash("after-test:"+test.title);}};'
+            WDIO_CONFIG='exports.config={runner:"local",hostname:"127.0.0.1",port:4723,path:"/wd/hub",specs:["tests/*.spec.js","tests/*.test.js"],maxInstances:1,bail:0,capabilities:[{platformName:"iOS","appium:automationName":"XCUITest","appium:bundleId":"'${{ env.APP_BUNDLE_ID }}'","appium:newCommandTimeout":300,"appium:noReset":true,"appium:forceAppLaunch":false,"appium:usePrebuiltWDA":true,"appium:wdaLocalPort":8100,"appium:showIOSLog":true,"appium:realDeviceLogger":"/usr/local/lib/node_modules/appium/node_modules/deviceconsole/deviceconsole"}],logLevel:"debug",waitforTimeout:120000,connectionRetryTimeout:30000,connectionRetryCount:3,services:[],framework:"mocha",reporters:["spec"],mochaOpts:{ui:"bdd",timeout:900000},before:async function(capabilities,specs,browser){const BUNDLE_ID="'${{ env.APP_BUNDLE_ID }}'";global.appCrashed=false;global.checkAppCrash=async(stage)=>{try{const state=await browser.queryAppState(BUNDLE_ID);console.log("["+stage+"] App state: "+state+" (4=foreground,3=background,1=not running)");if(state<3){console.error("\\n🛑 APP CRASHED at "+stage+"! State="+state);console.error("Check device logs for BareKit/native errors.\\n");global.appCrashed=true;process.exit(1);}return state;}catch(e){console.log("["+stage+"] queryAppState error: "+e.message);return-1;}};console.log("Checking initial app state...");await global.checkAppCrash("startup");console.log("Waiting for app to initialize...");await browser.pause(5000);await global.checkAppCrash("after-pause");const initText=await browser.$("-ios predicate string:label CONTAINS \\"INITIALIZED\\"");await initText.waitForDisplayed({timeout:60000});await global.checkAppCrash("after-init");console.log("App initialized, clicking Run Automated Tests...");const button=await browser.$("-ios predicate string:label CONTAINS \\"Run Automated Tests\\"");await button.waitForDisplayed({timeout:15000});await button.click();console.log("Button clicked!");await browser.pause(5000);await global.checkAppCrash("after-click");},afterTest:async function(test,context,{error}){if(global.appCrashed)return;await global.checkAppCrash("after-test:"+test.title);}};'
           fi
 
           # Base64 encode the wdio config to safely embed in YAML

.github/workflows/integration-test-qvac-lib-infer-whispercpp.yml

@@ -1,4 +1,4 @@
-name: Integration Tests (Whispercpp)
+name: Run Integration Tests
 
 on:
   workflow_dispatch:
@@ -46,7 +46,7 @@ jobs:
           - os: macos-14-xlarge
             platform: darwin
             arch: arm64
-          - os: windows-2022
+          - os: ai-run-windows-gpu
             platform: win32
             arch: x64
 
@@ -68,7 +68,8 @@ jobs:
           repository: ${{ inputs.repository || github.repository }}
           token: ${{ secrets.PAT_TOKEN }}
 
-      - name: Configure scoped registry for @tetherto and @qvac packages
+      - name: Configure scoped registry for @tetherto and @qvac packages (Unix)
+        if: ${{ matrix.platform != 'win32' }}
         working-directory: ${{ env.WORKDIR }}
         env:
           GPR_TOKEN: ${{ secrets.GITHUB_TOKEN }}
@@ -80,6 +81,7 @@ jobs:
           set -eu
           echo "Writing .npmrc for dual registry install…"
           cat > .npmrc <<NPMRC
+          always-auth=true
           registry=https://registry.npmjs.org/
           @qvac:registry=https://registry.npmjs.org/
           @tetherto:registry=https://npm.pkg.github.com/
@@ -95,12 +97,36 @@ jobs:
             git config --global url."https://${{ github.token }}:@github.com/".insteadOf "https://github.com/"
           fi
 
+      - name: Configure scoped registry for @tetherto and @qvac packages (Windows)
+        if: ${{ matrix.platform == 'win32' }}
+        working-directory: ${{ env.WORKDIR }}
+        env:
+          GPR_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          NPM_TOKEN: ${{ secrets.NPM_TOKEN }}
+          GIT_PAT: ${{ secrets.PAT_TOKEN }}
+        shell: powershell
+        run: |
+          Write-Host "Configuring scoped registry for @tetherto and @qvac packages..."
+          $npmrc = @"
+          always-auth=true
+          registry=https://registry.npmjs.org/
+          @qvac:registry=https://registry.npmjs.org/
+          @tetherto:registry=https://npm.pkg.github.com/
+          //registry.npmjs.org/:_authToken=$env:NPM_TOKEN
+          //npm.pkg.github.com/:_authToken=$env:GPR_TOKEN
+          "@
+          $npmrc | Out-File -FilePath .npmrc -Encoding utf8
+          if ($env:GIT_PAT) {
+            git config --global url."https://$($env:GIT_PAT):@github.com/".insteadOf "https://github.com/"
+          } else {
+            git config --global url."https://${{ github.token }}:@github.com/".insteadOf "https://github.com/"
+          }
+
       - name: Install NPM dependencies
         working-directory: ${{ env.WORKDIR }}
-        shell: bash
         run: |
           npm install
-          npm install -g bare bare-make
+          npm install -g bare@1.26.0 bare-make
 
       - name: Download prebuilds from artifact
         if: ${{ !inputs.prebuild_package }}
@@ -109,15 +135,33 @@ jobs:
           path: ${{ env.WORKDIR }}/prebuilds
           merge-multiple: true
 
-      - name: Download prebuilds from package
-        if: ${{ inputs.prebuild_package }}
+      - name: Download prebuilds from package (Unix)
+        if: ${{ inputs.prebuild_package && matrix.platform != 'win32' }}
         working-directory: ${{ env.WORKDIR }}
         shell: bash
         run: |
           mkdir -p prebuilds
-          npm pack "${{ inputs.prebuild_package }}" --pack-destination /tmp
+          npm pack ${{ inputs.prebuild_package }} --pack-destination /tmp
           tar -xzf /tmp/*.tgz -C /tmp
           cp -r /tmp/package/prebuilds/* prebuilds/
+          find prebuilds -type f -name 'tetherto__*' | while read f; do
+            mv "$f" "${f/tetherto__/qvac__}"
+          done
+
+      - name: Download prebuilds from package (Windows)
+        if: ${{ inputs.prebuild_package && matrix.platform == 'win32' }}
+        working-directory: ${{ env.WORKDIR }}
+        shell: powershell
+        run: |
+          New-Item -ItemType Directory -Force -Path prebuilds | Out-Null
+          npm pack ${{ inputs.prebuild_package }} --pack-destination $env:TEMP
+          $tgz = Get-ChildItem "$env:TEMP\*.tgz" | Select-Object -First 1
+          tar -xzf $tgz.FullName -C $env:TEMP
+          Copy-Item -Path "$env:TEMP\package\prebuilds\*" -Destination prebuilds -Recurse -Force
+          Get-ChildItem -Path prebuilds -Recurse -File -Filter 'tetherto__*' | ForEach-Object {
+            $newName = $_.Name -replace 'tetherto__', 'qvac__'
+            Rename-Item -Path $_.FullName -NewName $newName
+          }
 
       - name: Install gh cli for ai-run-linux-gpu
         if: matrix.os == 'ai-run-linux-gpu'
@@ -155,18 +199,36 @@ jobs:
         run: |
           brew install --quiet openblas lapack fftw
 
-      - name: Run integration test
+      - name: Run integration test (Unix)
+        if: ${{ matrix.platform != 'win32' }}
         working-directory: ${{ env.WORKDIR }}
         shell: bash
-        run: |
-          npm run test:integration
+        run: npm run test:integration
         env:
           GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
 
-      - name: Print run state
+      - name: Run integration test (Windows)
+        if: ${{ matrix.platform == 'win32' }}
+        working-directory: ${{ env.WORKDIR }}
+        shell: powershell
+        run: npm run test:integration
+        env:
+          GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
+
+      - name: Print run state (Unix)
+        if: ${{ matrix.platform != 'win32' }}
         working-directory: ${{ env.WORKDIR }}
         shell: bash
         run: |
           ls -la prebuilds || true
           tree prebuilds || true
         continue-on-error: true
+
+      - name: Print run state (Windows)
+        if: ${{ matrix.platform == 'win32' }}
+        working-directory: ${{ env.WORKDIR }}
+        shell: powershell
+        run: |
+          Get-ChildItem prebuilds -ErrorAction SilentlyContinue
+          tree prebuilds /F 2>$null
+        continue-on-error: true

packages/qvac-lib-infer-whispercpp/addon/src/model-interface/whisper.cpp/WhisperModel.cpp

@@ -68,13 +68,6 @@ void WhisperModel::load() {
       throw std::runtime_error("Failed to initialize Whisper context");
     }
 
-    state_.reset(whisper_init_state(ctx_.get()));
-    if (state_ == nullptr) {
-      QLOG(
-          qvac_lib_inference_addon_cpp::logger::Priority::ERROR,
-          "Failed to initialize Whisper state");
-      throw std::runtime_error("Failed to initialize Whisper state");
-    }
     is_loaded_ = true;
     QLOG(
         qvac_lib_inference_addon_cpp::logger::Priority::INFO,
@@ -238,9 +231,8 @@ void WhisperModel::warmup() {
   params.new_segment_callback_user_data = nullptr;
 
   // Run warmup inference to "heat up" the model
-  whisper_full_with_state(
+  whisper_full(
       ctx_.get(),
-      state_.get(),
       params,
       silentAudio.data(),
       static_cast<int>(silentAudio.size()));
@@ -282,12 +274,8 @@ void WhisperModel::process(const Input& input) {
   params.new_segment_callback = onNewSegment;
   params.new_segment_callback_user_data = &ud;
 
-  int result = whisper_full_with_state(
-      ctx_.get(),
-      state_.get(),
-      params,
-      input.data(),
-      static_cast<int>(input.size()));
+  int result = whisper_full(
+      ctx_.get(), params, input.data(), static_cast<int>(input.size()));
 
   const auto t1 = std::chrono::steady_clock::now();
   totalWallMs_ += std::chrono::duration<double, std::milli>(t1 - t0).count();
@@ -371,10 +359,7 @@ bool WhisperModel::configContextIsChanged(
   return false;
 }
 
-void WhisperModel::resetContext() {
-  ctx_.reset();
-  state_.reset();
-}
+void WhisperModel::resetContext() { ctx_.reset(); }
 
 void WhisperModel::setConfig(const WhisperConfig& config) {
   bool contextChanged = configContextIsChanged(cfg_, config);

packages/qvac-lib-infer-whispercpp/addon/src/model-interface/whisper.cpp/WhisperModel.hpp

@@ -121,16 +121,7 @@ class WhisperModel {
     }
   };
 
-  struct WhisperStateDeleter {
-    void operator()(whisper_state* state) const noexcept {
-      if (state != nullptr) {
-        whisper_free_state(state);
-      }
-    }
-  };
-
   std::unique_ptr<whisper_context, WhisperContextDeleter> ctx_{nullptr};
-  std::unique_ptr<whisper_state, WhisperStateDeleter> state_{nullptr};
   bool stream_ended_ = false;
   bool is_loaded_ = false;
   bool is_warmed_up_ = false;

packages/qvac-lib-infer-whispercpp/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@qvac/transcription-whispercpp",
-  "version": "0.3.17",
+  "version": "0.3.18",
   "description": "transcription addon for qvac",
   "addon": true,
   "engines": {

packages/qvac-lib-infer-whispercpp/release-notes/v0.3.18.md

@@ -0,0 +1,25 @@
+# QVAC Transcription Whisper Addon v0.3.18 Release Notes
+
+This release updates the whisper.cpp integration to use the latest API and improves Windows platform support for CI/CD workflows.
+
+## Features
+
+### Enhanced Windows Platform Support
+
+The integration test workflow now includes comprehensive Windows support with PowerShell-specific configurations. This includes separate registry configuration steps for Unix and Windows platforms, ensuring proper package authentication across all environments. The workflow now uses the `ai-run-windows-gpu` runner for better GPU-accelerated testing on Windows.
+
+### Prebuild Package Renaming Support
+
+Added automatic renaming of prebuild packages from `tetherto__*` to `qvac__*` format during the package download process. This ensures consistency across all platforms and simplifies the build artifact management.
+
+## Bug Fixes
+
+### Whisper.cpp API Compatibility
+
+Updated the integration with whisper.cpp to use the new 4-parameter API for `whisper_full()`. The previous 5-parameter version that included an explicit state parameter has been deprecated. The state is now managed internally by the whisper.cpp context, simplifying the code and removing unnecessary state management overhead.
+
+Removed the obsolete `whisper_state` member variable and related initialization code, as the new whisper.cpp API handles state management automatically through the context object.
+
+## Other
+
+Updated the integration test workflow to use a specific version of `bare@1.26.0` for better build consistency. Added the `always-auth=true` configuration for npm registry authentication to improve reliability when downloading private packages.