visual c++ 创建Win32工程调用windows API 做语音识别。查找了很多资料，但是很少用Win32的，国外倒是有很多人用C#调用Windows API 做语音识别。很多结合语音识别与word结合，把识别的语音写在word文档上，由于我不是C#喜好者，所以没有查找实现那些的source code。

代码如下：

// SpeechToTextTest2.cpp : Defines the entry point for the application.
#include "stdafx.h"
#include "SpeechToTextTest2.h"
#include<Windows.h>
#include<WinUser.h>
#include <sphelper.h>
#include <string>
//#include "Resource.h"
#define WM_RECOEVENT    WM_USER+1
#define ID_START_RECOG 13
BOOL CALLBACK DlgProc(HWND hWnd, UINT Message, WPARAM wParam, LPARAM lParam);
void LaunchRecognition(HWND hWnd);
void HandleEvent(HWND hWnd);
WCHAR *ExtractInput(CSpEvent event);
void CleanupSAPI();
CComPtr<ISpRecognizer> g_cpEngine;
CComPtr<ISpRecoContext> g_cpRecoCtx;
CComPtr<ISpRecoGrammar> g_cpRecoGrammar;
WCHAR *lpszBuffer;
int WINAPI WinMain(HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR lpCmdLine, int nShowCmd)
	// allocating memory for buffer this buffer is used to store
	// the text during the speech recognition process
	lpszBuffer = new WCHAR[MAX_PATH];
	lpszBuffer[0] = 0;
	DialogBox(hInstance, MAKEINTRESOURCE(IDD_DIALOG12), NULL, DlgProc);
	// freeing the memory that was allocated for the buffer
	delete[] lpszBuffer;
	return 0;
BOOL CALLBACK DlgProc(HWND hWnd, UINT Message, WPARAM wParam, LPARAM lParam)
	switch (Message)
	case WM_RECOEVENT:
		HandleEvent(hWnd);
		break;
	case WM_COMMAND:
		switch (LOWORD(wParam))
		case IDC_BUTTON1:
			LaunchRecognition(hWnd);
			break;
		break;
	case WM_CLOSE:
		CleanupSAPI();
		EndDialog(hWnd, 0);
		break;
	default:
		return FALSE;
	return TRUE;
void LaunchRecognition(HWND hWnd)
	if (FAILED(::CoInitialize(NULL)))
		throw std::string("Unable to initialise COM objects");
	ULONGLONG ullGramId = 1;
	HRESULT hr = g_cpEngine.CoCreateInstance(CLSID_SpSharedRecognizer);
	if (FAILED(hr))
		throw std::string("Unable to create recognition engine");
	hr = g_cpEngine->CreateRecoContext(&g_cpRecoCtx);
	if (FAILED(hr))
		throw std::string("Failed command recognition");
	hr = g_cpRecoCtx->SetNotifyWindowMessage(hWnd, WM_RECOEVENT, 0, 0);
	if (FAILED(hr))
		throw std::string("Unable to select notification window");
	const ULONGLONG ullInterest = SPFEI(SPEI_SOUND_START) | SPFEI(SPEI_SOUND_END) |
		SPFEI(SPEI_PHRASE_START) | SPFEI(SPEI_RECOGNITION) |
		SPFEI(SPEI_FALSE_RECOGNITION) | SPFEI(SPEI_HYPOTHESIS) |
		SPFEI(SPEI_INTERFERENCE) | SPFEI(SPEI_RECO_OTHER_CONTEXT) |
		SPFEI(SPEI_REQUEST_UI) | SPFEI(SPEI_RECO_STATE_CHANGE) |
		SPFEI(SPEI_PROPERTY_NUM_CHANGE) | SPFEI(SPEI_PROPERTY_STRING_CHANGE);
	hr = g_cpRecoCtx->SetInterest(ullInterest, ullInterest);
	if (FAILED(hr))
		throw std::string("Failed to create interest");
	hr = g_cpRecoCtx->CreateGrammar(ullGramId, &g_cpRecoGrammar);
	if (FAILED(hr))
		throw std::string("Unable to create grammar");
	hr = g_cpRecoGrammar->LoadDictation(0, SPLO_STATIC);
	if (FAILED(hr))
		throw std::string("Failed to load dictation");
	hr = g_cpRecoGrammar->SetDictationState(SPRS_ACTIVE);
	if (FAILED(hr))
		throw std::string("Failed setting dictation state");
void HandleEvent(HWND hWnd)
	CSpEvent event;
	WCHAR  *pwszText;
	// Loop processing events while there are any in the queue
	while (event.GetFrom(g_cpRecoCtx) == S_OK)
		switch (event.eEventId)
		case SPEI_HYPOTHESIS:
			pwszText = ExtractInput(event);
			MessageBoxW(NULL, pwszText, L"text", MB_ICONERROR);
			wcscat(lpszBuffer, pwszText);
			wcsncat(lpszBuffer, L"\r\n", 2);
			SetDlgItemTextW(hWnd, IDC_EDIT1, lpszBuffer);
		break;
WCHAR *ExtractInput(CSpEvent event)
	HRESULT                   hr = S_OK;
	CComPtr<ISpRecoResult>    cpRecoResult;
	SPPHRASE                  *pPhrase;
	WCHAR                     *pwszText;
	cpRecoResult = event.RecoResult();
	hr = cpRecoResult->GetPhrase(&pPhrase);
	if (SUCCEEDED(hr))
		if (event.eEventId == SPEI_FALSE_RECOGNITION)
			pwszText = L"False recognition";
			//MessageBoxW(NULL, pwszText, L"text", MB_ICONERROR);
			// Get the phrase's entire text string, including replacements.
			hr = cpRecoResult->GetText(SP_GETWHOLEPHRASE, SP_GETWHOLEPHRASE, TRUE, &pwszText, NULL);
	CoTaskMemFree(pPhrase);
	return pwszText;
void CleanupSAPI()
	if (g_cpRecoGrammar)
		g_cpRecoGrammar.Release();
	if (g_cpRecoCtx)
		g_cpRecoCtx->SetNotifySink(NULL);
		g_cpRecoCtx.Release();
	if (g_cpEngine)
		g_cpEngine.Release();
	CoUninitialize();
运行结果如下：
 
C++ Speech Recognition：https://www.codeproject.com/Tips/784140/Cplusplus-Speech-Recognition
 
点进去看这个文章后，你会发现有一堆问都文件问题的人，无法编译通过，因为源工程已经不存在了，.cpp而没有头文件部分。经过一番摸索，我实现了如上。
 
博客里面也只有.cpp部分，由于csdn下载文件要积分。所以就不上传了源工程了。
 
如果觉得有参考价值、需要源文件的同志，可以email我，邮箱：761577651@qq.com
 
也可以在下方留言：留言比较快！向那些迟发的邮件致歉！
 
不缺积分的可以点击链接下载：https://download.csdn.net/download/thecentry/10859369
 
还要提醒的是，我只是实现了，源博客APP的实现问题。
                    visual c++ 创建Win32工程调用windows API 做语音识别。查找了很多资料，但是很少用Win32的，国外倒是有很多人用C#调用Windows API 做语音识别。很多结合语音识别与word结合，把识别的语音写在word文档上，由于我不是C#喜好者，所以没有查找实现那些的source code。代码如下：// SpeechToTextTest2.cpp : Define...
语音特征
MFCC：Mel Frequency Cepstral Ceofficient，目前最主流的语音信号特征提取方式，相比ceptrum的流程，主要是增加了mel滤波，另外用DCT替换了IFFT。
Fbank：亦称MFSC(log mel-frequency spectral Coefficients)，特征的提取方法就是相当于MFCC去掉最后一步的离散余弦变换，跟MFCC特征相比，Fbank特征保留了更多的原始语音数据。
声学模型：