Fix/whisper silence hallucination

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

@@ -1,4 +1,4 @@
-name: Integration Tests (Whispercpp)
+name: Run Integration Tests
 
 on:
   workflow_dispatch:
@@ -10,7 +10,7 @@ on:
   workflow_call:
     inputs:
       ref:
-        description: "ref"
+        description: 'ref'
         type: string
       repository:
         type: string
@@ -21,19 +21,16 @@ jobs:
     continue-on-error: true
     runs-on: ${{ matrix.os }}
     name: ${{ matrix.platform }}-${{ matrix.arch }}-integration-tests
-
     env:
       GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
-      WORKDIR: packages/qvac-lib-infer-whispercpp
-
     permissions:
       contents: read
       packages: read
-
     strategy:
       fail-fast: false
       matrix:
         include:
+          # Run Linux x64 tests on both Ubuntu 22.04 and 24.04 to ensure compatibility
           - os: ubuntu-22.04
             platform: linux
             arch: x64
@@ -46,7 +43,7 @@ jobs:
           - os: macos-14-xlarge
             platform: darwin
             arch: arm64
-          - os: windows-2022
+          - os: ai-run-windows-gpu
             platform: win32
             arch: x64
 
@@ -68,8 +65,8 @@ jobs:
           repository: ${{ inputs.repository || github.repository }}
           token: ${{ secrets.PAT_TOKEN }}
 
-      - name: Configure scoped registry for @tetherto and @qvac packages
-        working-directory: ${{ env.WORKDIR }}
+      - name: Configure scoped registry for @tetherto and @qvac packages (Unix)
+        if: ${{ matrix.platform != 'win32' }}
         env:
           GPR_TOKEN: ${{ secrets.GITHUB_TOKEN }}
           NPM_TOKEN: ${{ secrets.NPM_TOKEN }}
@@ -80,6 +77,7 @@ jobs:
           set -eu
           echo "Writing .npmrc for dual registry install…"
           cat > .npmrc <<NPMRC
+          always-auth=true
           registry=https://registry.npmjs.org/
           @qvac:registry=https://registry.npmjs.org/
           @tetherto:registry=https://npm.pkg.github.com/
@@ -95,29 +93,82 @@ jobs:
             git config --global url."https://${{ github.token }}:@github.com/".insteadOf "https://github.com/"
           fi
 
+      - name: Configure scoped registry for @tetherto and @qvac packages (Windows)
+        if: ${{ matrix.platform == 'win32' }}
+        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 }}
         uses: actions/download-artifact@v4
         with:
-          path: ${{ env.WORKDIR }}/prebuilds
+          path: prebuilds
           merge-multiple: true
 
-      - name: Download prebuilds from package
-        if: ${{ inputs.prebuild_package }}
-        working-directory: ${{ env.WORKDIR }}
+      - name: Download prebuilds from package (Unix)
+        if: ${{ inputs.prebuild_package && matrix.platform != 'win32' }}
         shell: bash
         run: |
           mkdir -p prebuilds
-          npm pack "${{ 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' }}
+        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 $newName
+          }
+
+      - name: Install gh cli for ai-run-linux-gpu
+        if: matrix.os == 'ai-run-linux-gpu'
+        shell: bash
+        run: |
+          (type -p wget >/dev/null || (sudo apt update && sudo apt-get install wget -y)) \
+            && sudo mkdir -p -m 755 /etc/apt/keyrings \
+                  && out=$(mktemp) && wget -nv -O$out https://cli.github.com/packages/githubcli-archive-keyring.gpg \
+                  && cat $out | sudo tee /etc/apt/keyrings/githubcli-archive-keyring.gpg > /dev/null \
+            && sudo chmod go+r /etc/apt/keyrings/githubcli-archive-keyring.gpg \
+            && echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/githubcli-archive-keyring.gpg] https://cli.github.com/packages stable main" | sudo tee /etc/apt/sources.list.d/github-cli.list > /dev/null \
+            && sudo apt update \
+            && sudo apt install gh -y
+          gh --version
+        continue-on-error: true
 
       - name: Install gh cli for ai-run-linux-gpu
         if: matrix.os == 'ai-run-linux-gpu'
@@ -155,18 +206,32 @@ jobs:
         run: |
           brew install --quiet openblas lapack fftw
 
-      - name: Run integration test
-        working-directory: ${{ env.WORKDIR }}
+      - name: Run integration test (Unix)
+        if: ${{ matrix.platform != 'win32' }}
         shell: bash
-        run: |
-          npm run test:integration
+        run: npm run test:integration
+        env:
+          GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
+
+      - name: Run integration test (Windows)
+        if: ${{ matrix.platform == 'win32' }}
+        shell: powershell
+        run: npm run test:integration
         env:
           GITHUB_TOKEN: ${{ secrets.PAT_TOKEN }}
 
-      - name: Print run state
-        working-directory: ${{ env.WORKDIR }}
+      - name: Print run state (Unix)
+        if: ${{ matrix.platform != 'win32' }}
         shell: bash
         run: |
           ls -la prebuilds || true
           tree prebuilds || true
         continue-on-error: true
+
+      - name: Print run state (Windows)
+        if: ${{ matrix.platform == 'win32' }}
+        shell: powershell
+        run: |
+          Get-ChildItem prebuilds -ErrorAction SilentlyContinue
+          tree prebuilds /F 2>$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()));
@@ -282,9 +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(
+  int result = whisper_full(
       ctx_.get(),
-      state_.get(),
       params,
       input.data(),
       static_cast<int>(input.size()));
@@ -373,7 +364,6 @@ bool WhisperModel::configContextIsChanged(
 
 void WhisperModel::resetContext() {
   ctx_.reset();
-  state_.reset();
 }
 
 void WhisperModel::setConfig(const WhisperConfig& 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.