Bugfix: fix use-after-free in GetMicsImpl

* Plumb beam search params into whisper cpp implementation
  (currently broken)

5 files changed, 28 insertions, 16 deletions

diff --git a/GUI/GUI/GUI/GUI.vcxproj b/GUI/GUI/GUI/GUI.vcxproj
index 6976c31..2d08d30 100644
--- a/GUI/GUI/GUI/GUI.vcxproj
+++ b/GUI/GUI/GUI/GUI.vcxproj
@@ -44,6 +44,7 @@
     <UseDebugLibraries>true</UseDebugLibraries>

     <PlatformToolset>v143</PlatformToolset>

     <CharacterSet>Unicode</CharacterSet>

+    <EnableASAN>false</EnableASAN>

   </PropertyGroup>

   <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">

     <ConfigurationType>Application</ConfigurationType>

diff --git a/GUI/GUI/GUI/WhisperCPP.cpp b/GUI/GUI/GUI/WhisperCPP.cpp
index ec8a6c2..6fd38b8 100644
--- a/GUI/GUI/GUI/WhisperCPP.cpp
+++ b/GUI/GUI/GUI/WhisperCPP.cpp
@@ -113,14 +113,14 @@ bool WhisperCPP::GetMics(std::vector<std::string>& mics) {
 		return false;

 	}

 

-	std::vector<std::unique_ptr<sCaptureDevice>> mics_raw;

+	std::vector<std::unique_ptr<MicInfo>> mics_raw;

 	if (!GetMicsImpl(mics_raw)) {

 		return false;

 	}

 

 	mics.clear();

 	for (const auto& raw_mic : mics_raw) {

-		mics.push_back(wcharToAsciiString(raw_mic->displayName));

+		mics.push_back(wcharToAsciiString(raw_mic->name.c_str()));

 	}

 

 	return true;

@@ -132,7 +132,7 @@ bool WhisperCPP::OpenMic(const int idx, Whisper::iAudioCapture*& stream) {
 		return false;

 	}

 

-	std::vector<std::unique_ptr<sCaptureDevice>> mics_raw;

+	std::vector<std::unique_ptr<MicInfo>> mics_raw;

 	if (!GetMicsImpl(mics_raw)) {

 		return false;

 	}

@@ -149,11 +149,11 @@ bool WhisperCPP::OpenMic(const int idx, Whisper::iAudioCapture*& stream) {
 	params.maxDuration = 3.0;

 	params.retainDuration = 1.5;

 	stream = nullptr;

-	HRESULT err = f_->openCaptureDevice(mics_raw[idx]->endpoint, params,

-		&stream);

+	HRESULT err = f_->openCaptureDevice(mics_raw[idx]->endpoint.c_str(),

+		params, &stream);

 	if (FAILED(err)) {

 		Log(out_, "Failed to open mic with idx {} ({}): {}\n", idx,

-			wcharToAsciiString(mics_raw[idx]->displayName),

+			wcharToAsciiString(mics_raw[idx]->name.c_str()),

 			hresultToString(err));

 		return false;

 	}

@@ -318,7 +318,9 @@ void WhisperCPP::Start(const AppConfig& c) {
 		ScopeGuard context_cleanup([context]() { context->Release(); });

 

 		Whisper::sFullParams wparams{};

-		context->fullDefaultParams(eSamplingStrategy::Greedy, &wparams);

+		context->fullDefaultParams(eSamplingStrategy::BeamSearch, &wparams);

+		wparams.beam_search.beam_width = 5;

+		wparams.beam_search.n_best = 5;

 		wparams.language = Whisper::makeLanguageKey("en");  // TODO(yum) use config

 		// This must be set to keep memory usage from growing without bound.

 		wparams.n_max_text_ctx = 100;

@@ -538,11 +540,11 @@ void WhisperCPP::StopCustomChatbox() {
 	Log(out_, "Done!\n");

 }

 

-bool WhisperCPP::GetMicsImpl(std::vector<std::unique_ptr<sCaptureDevice>>& mics) {

+bool WhisperCPP::GetMicsImpl(std::vector<std::unique_ptr<MicInfo>>& mics) {

 	pfnFoundCaptureDevices dev_cb = [](int len, const sCaptureDevice* buf, void* pv)->HRESULT __stdcall {

-		auto mics = static_cast<std::vector<std::unique_ptr<sCaptureDevice>>*>(pv);

+		auto mics = static_cast<std::vector<std::unique_ptr<MicInfo>>*>(pv);

 		for (int i = 0; i < len; i++) {

-			mics->push_back(std::make_unique<sCaptureDevice>(buf[i]));

+			mics->push_back(std::make_unique<MicInfo>(buf[i].displayName, buf[i].endpoint));

 		}

 		return S_OK;

 	};

diff --git a/GUI/GUI/GUI/WhisperCPP.h b/GUI/GUI/GUI/WhisperCPP.h
index 7b3f17d..d58a671 100644
--- a/GUI/GUI/GUI/WhisperCPP.h
+++ b/GUI/GUI/GUI/WhisperCPP.h
@@ -46,8 +46,16 @@ public:
 	void StopCustomChatbox();

 

 private:

+	struct MicInfo {

+		MicInfo(const wchar_t* n, const wchar_t* e)

+			: name(n), endpoint(e)

+		{}

+

+		std::wstring name;

+		std::wstring endpoint;

+	};

 	bool GetMicsImpl(

-		std::vector<std::unique_ptr<Whisper::sCaptureDevice>>& mics);

+		std::vector<std::unique_ptr<MicInfo>>& mics);

 

 	wxTextCtrl* out_;

 	Whisper::iMediaFoundation* f_;

diff --git a/GUI/README.md b/GUI/README.md
index ea28124..dbbc9be 100644
--- a/GUI/README.md
+++ b/GUI/README.md
@@ -8,17 +8,18 @@ $ git submodule init
 $ git submodule update
 ```
 3. Execute Libraries/fetch.ps1.
-4. Open Libraries/wx/build/msw/wx\_vc17.sln with Visual Studio 2022.
-5. Select every project in the Solution Explorer except for _custom_build.
+4. Open `Libraries/wx/build/msw/wx_vc17.sln` with Visual Studio 2022.
+5. Select every project in the Solution Explorer except for `_custom_build`.
 6. Right click, select Properties, go to C/C++, Code Generation, and set
    Runtime Library to Multi-threaded (/MT). Make sure this applies to the
    configuration x64/Release.
 7. Build x64/Release.
   1. The build configuration is in the top. By default it's probably Debug/x64.
   2. To build: ctrl+shift+B
-8. Open GUI/GUI.sln with Visual Studio 2022.
-9. Build x64/Release.
-10. Run package.ps1 from powershell.
+8. Follow TaSTT-Whisper README and build it as x64/Release.
+9. Open GUI/GUI.sln with Visual Studio 2022.
+10. Build x64/Release.
+11. Run package.ps1 from powershell.
 
 ## High level design
 
diff --git a/TaSTT-Whisper b/TaSTT-Whisper
-Subproject 1136acfc365f357d2df13a263714e8ae0614c4f
+Subproject 8050fba80e08dc2d107944fb20da7028dc73d05