MicArrayEchoCancellation Walkthrough: C++–1
MicArrayEchoCancellationWalkthrough: C++
Capturing Audio Streams with Acoustic Echo Cancellation andBeamforming
About This Walkthrough In the Kinect™ for Windows® Software Development Kit (SDK), the MicArrayEchoCancellation sample shows how to capture an audio stream from the microphone array of the Kinect for Xbox 360®sensor by using the MSRKinectAudio Microsoft DirectX® media object (DMO) in a Microsoft DirectShow® graph. This document provides a walkthrough review of the MicArrayEchoCancellation sample.
Resources For a complete list of documentation for the Kinect for Windows SDK Beta, plus related reference and links to the online forums, see the beta SDK website at:
Contents
Introduction
Program Description
Create and Configure the MSRKinectAudio DMO
Select the Kinect Sensor’s Microphone Array
Enumerate the Device Index
Determine the Device Index
Record the Captured Stream and Determine the Source Direction
Set Up the Data Buffer
Set the Output Format
Allocate Resources and the Output Buffer
Capture the Audio Stream and Determine Source Direction
License: The Kinect for Windows SDK Beta is licensed for non-commercial use only. By installing, copying, or otherwise using the beta SDK, you agree to be bound by the terms of its license. Read the license.
Disclaimer: This document is provided “as-is”. Information and views expressed in this document, including URL and other Internet Web site references, may change without notice. You bear the risk of using it.
This document does not provide you with any legal rights to any intellectual property in any Microsoft product. You may copy and use this document for your internal, reference purposes.
© 2011 Microsoft Corporation. All rights reserved.
Microsoft, DirectShow, DirectX, Kinect, MSDN, Windows, and Windows Media are trademarks of the Microsoft group of companies. All other trademarks are property of their respective owners.
Introduction
The audio component of the Kinect™ for Xbox 360®sensor is a four-element linear microphone array. An array provides some significant advantages over a single microphone, including more sophisticated acoustic echo cancellation and noise suppression, and the ability to determine the direction of a sound source.
The primary way for C++ applications to access the Kinect sensor’s microphone array is through the MSRKinectAudio Microsoft® DirectX® media object (DMO). A DMO is a standard COM object that can be incorporated into a Microsoft DirectShow® graph or a Microsoft Media Foundation topology. The Kinect for Windows®Software Development Kit (SDK)Beta includes an extended version of the Windows microphone array DMO—referred to here as the MSRKinectAudio DMO—to support the Kinect microphone array.
The MSRKinectAudio DMO supports all the standard microphone array functionality, which includes:
- Acoustic echo cancellation (AEC)
- Microphone array processing (MicArray)
- Noise suppression (NS)
- Automatic gain control (AGC)
- Voice activity detection (VAD)
- Sound source localization, which identifies the direction of the source in the horizontal plane
- Beamforming, which allows the array to function as a steerable directional microphone.
The DMO supports 11 beams, with fixed directions that range from -50 to+50 degrees in 10degree increments.
For more information on the standard microphone array, see “Microphone Array Support in Windows Vista” and “How to Build and Use Microphone Arrays for Windows Vista” on the Microsoft Developer Network (MSDN®) website.
Although the internal details for MSRKinectAudio DMO are different, you use it in much the same way as the standard microphone array DMO, with the following exceptions. The MSRKinectAudio DMO:
- Has its own class identifier (CLSID)—CLSID_CMSRMSRKinectAudio.
- Exposes sound source localization functionality through a new interface—ISoundSourceLocalizer.
- Supports an additional microphone array mode—adaptive beamforming)—which uses an internal source localizer to automatically determine the beam direction.
The MicArrayEchoCancellation sample shows how to capture an audio stream from the Kinect sensor’s microphone array by polling the MSRKinectAudio DMO in source mode. The application uses AEC to record a high-quality audio stream and beam-forming to determine the direction to the sound source. The DMO can also be used with a Microsoft Media Foundation topology. For an example, see “MFAudioFilter Walkthrough: C++ Sample” on the beta SDK website.
Note DirectShow is COM-based, and this document assumes that you are familiar with how to use COM objects and interfaces. You do not need to know how to implement COM objects. For the basics of how to use COM objects, see “Programming DirectX with COM” on the MSDN website.That MSDN topic is written for DirectX programmers, but the basic principles apply to all COM-based applications.
Program Description
MicArrayEchoCancellationis installed with the Kinect for Windows Software Development Kit (SDK) Beta samples in %KINECTSDK_DIR%\Samples\KinectSDKSamples.zip..MicArrayEchoCancellation is a C++ console application that is implemented in MicArrayEchoCancellation.cpp.
The basic program flow is as follows:
1.Create and configure the MSRKinectAudio DMO.
2.Enumerate the available capture devices and select the Kinect sensor’s microphone array.
3.Record 10 seconds of audio stream and determine the source direction as the capture process progresses.
To run MicArrayEchoCancellation, start MicArrayEchoCancellation.exe and follow the instructions in the console window.
Tip Before attempting to capture audio from the microphone array, you must be actively streaming to the audio render device that is specified for the DMO—typically the system’s speakers. Otherwise, the MSRKinectAudio DMO fails. AEC is designed to cancel interfering sounds, so there must be something to cancel.The simplest solution is to start playing a tune on Windows Media® Player before you run the application. The Libraries\Music\Sample Music folder on your Windows PC contains some sample music files.
The following is a lightly edited version of the output from a MicArrayEchoCancellation session, where the sound source moved from side to side as capture progressed:
Start a song in Windows Media Player and then press any key to start recording (echo cancellation processing expects speakers to be producing sound).
Recording using DMO
AEC-MicArray is running ... Press "s" to stop
Position: -0.051290 Confidence: 1.000000 Beam Angle = 0.0000000
Sound output was written to file: C:\KDK\Samples\Audio\MicArrayEchoCancellation\CPP\AECout.wav
The recording process uses beamforming, whichcreates a single directional channel from the four microphones in 16-kHz, 16-bit mono pulse code modulation (PCM) format. The channel is oriented to one of the 11 beam directions. MicArrayEchoCancellation uses adaptive beamforming, which automatically selects the beam that is closest to the source direction.
You can use the captured stream for many purposes. MicArrayEchoCancellation simply writes the captured audio stream to AECout.wav—which is a .wav file that can be played with Windows Media Player.
The rest of this document is a walkthrough of the MicArrayEchoCancellation sample. It describes all the sample’s functionality except for writing the capture stream to a .wav file. For details on that process, see the sample code.
Note This document includes code excerpts, most of which have been edited for brevity and readability. In particular, most routine error-correction code has been removed. For the complete code, see the MicArrayEchoCancellation sample. Hyperlinks in this walkthrough refer to content on the MSDN website.
Create and Configure the MSRKinectAudio DMO
The application’s entry point—_tmain—manages the overall program execution, with private methods handling most details. The first step is to create and configure an instance of the MSRKinectAudio DMO, as follows:
#include "MSRKinectAudio.h"
...
int__cdecl _tmain(int argc, const TCHAR ** argv)
{
HRESULT hr = S_OK;
CoInitialize(NULL);
int iMicDevIdx = -1;
IMediaObject* pDMO = NULL;
IPropertyStore* pPS = NULL;
...
SetPriorityClass (GetCurrentProcess(), HIGH_PRIORITY_CLASS);
CoCreateInstance(CLSID_CMSRKinectAudio,
NULL, CLSCTX_INPROC_SERVER,
IID_IMediaObject, (void**)&pDMO);
pDMO->QueryInterface(IID_IPropertyStore, (void**)&pPS);
PROPVARIANT pvSysMode;
PropVariantInit(&pvSysMode);
pvSysMode.vt = VT_I4;
pvSysMode.lVal = (LONG)(4);
pPS->SetValue(MFPKEY_WMAAECMA_SYSTEM_MODE, pvSysMode);
PropVariantClear(&pvSysMode);
...
}
In addition to standard header files, the file includes the beta SDK header file, MSRKinectAudio.h. This file contains the DMO’s globally unique identifiers (GUIDs) and interface declarations for the MSRKinectAudio DMO.
Before creating the DMO, MicArrayEchoCancellation calls the SetPriorityClassfunction to set the process’s priority to HIGH_PRIORITY_CLASS. This helps ensure that the microphone is not preempted during the capture process.
MicArrayEchoCancellation calls the CoCreateInstancefunction to create an instance of the DMO and obtain its IMediaObject interface, which supports a set of methods that let you control the DMO. A DMO supports a property store—which you access through an IPropertyStore interface—that contains a collection of key-value pairs. You configure a DMO by setting the appropriate keys in the object’s property store.
DMO property store keys are GUIDs that are defined by Media Foundation. The associated values are packaged as a PROPVARIANT structure, which must be initialized by calling the PropVariantInitfunction. You then assign values to the following two members:
- PROPVARIANT.vt, a VARENUM value that specifies the data type.
For example, a 4-byte unsigned integer corresponds to a VT_I4 data type.
- A value member, whose name depends on vt.
For VT_I4, the value member is PROPVARIANT.lVal.
MicArrayEchoCancellation calls the DMO’s QueryInterface method to obtain an IPropertyStore pointer, and then calls the IPropertyStore::SetValue method to specify the system mode. The key is MFPKEY_WMAAECMA_SYSTEM_MODE, and four modes are available, each of which has a corresponding VT_I4 value as shown in the following table.
Mode / ValueSingle-channel with AEC / 0
Microphone array / 2
Microphone array with AEC / 4
Single-channel with automatic gain control / 5
MicArrayEchoCancellation uses the array with AEC, so the value is set to 4.
Note MicArrayEchoCancellation uses the MSRKinectAudio DMO’s default microphone array mode—which enables adaptive beamforming. So MicArrayEchoCancellationdoes not explicitly set MFPKEY_WMAAECMA_FEATR_MICARR_MODE.
Select the Kinect Sensor’s Microphone Array
Because a system can have more than one active microphone, MicArrayEchoCancellation must specify which microphone stream is to be captured by setting the MFPKEY_WMAAECMA_DEVICE_INDEXES key to the appropriate device index. To determine the index, MicArrayEchoCancellation enumerates the available microphones, determines which one is the Kinect sensor’s microphone array, and uses that index to set the value, as follows:
int__cdecl _tmain(int argc, const TCHAR ** argv)
{
...
hr = GetMicArrayDeviceIndex(&iMicDevIdx);
PROPVARIANT pvDeviceId;
PropVariantInit(&pvDeviceId);
pvDeviceId.vt = VT_I4;
pvDeviceId.lVal =
(unsignedlong)(iSpkDevIdx<16) |
(unsignedlong)(0x0000ffff & iMicDevIdx);
pPS->SetValue(MFPKEY_WMAAECMA_DEVICE_INDEXES, pvDeviceId);
PropVariantClear(&pvDeviceId);
...
}
The value of MFPKEY_WMAAECMA_DEVICE_INDEXES is a 32-bit integer that contains both speaker and microphone indices:
- The speaker index is in the upper 2 bytes.
- The microphone index is in the lower 2 bytes.
The private GetMicArrayDeviceIndex method enumerates the available capture devices and determines the Kinect sensor’s microphone array device index.
Enumerate the Device Index
The first step in enumerating the device index is to enumerate the available capture devices, as follows:
HRESULT GetMicArrayDeviceIndex(int *piDevice)
{
HRESULT hr = S_OK;
UINT index, dwCount;
IMMDeviceEnumerator* spEnumerator;
IMMDeviceCollection* spEndpoints;
*piDevice = -1;
CoCreateInstance(__uuidof(MMDeviceEnumerator),
NULL, CLSCTX_ALL, __uuidof(IMMDeviceEnumerator),
(void**)&spEnumerator);
spEnumerator->EnumAudioEndpoints(eCapture,
DEVICE_STATE_ACTIVE,
&spEndpoints);
...
}
GetMicArrayDeviceIndex:
1.Creates a device enumerator object, and gets a pointer to its IMMDeviceEnumerator interface.
2.Enumerates the system’s capture devices by calling the enumerator object’s IMMDeviceEnumerator::EnumAudioEndpoints method, which enumerates the specified types of audio endpoints.
The EnumAudioEndpoints parameter values are as follows:
- A value from the EDataFlow enumeration that indicates the device type.
eCapture directs EnumAudioEndpoints to enumerate only capture devices.
- A DEVICE_STATE_XXX constant that specifies which device states to enumerate.
DEVICE_STATE_ACTIVE directs EnumAudioEndpoints to enumerate only active devices.
- The address of an IMMDeviceCollection interface pointer that contains the enumerated capture devices.
Determine the Device Index
GetMicArrayDeviceIndex then determines the Kinect sensor’s microphone array device index, as follows:
HRESULT GetMicArrayDeviceIndex(int *piDevice)
{
...
spEndpoints->GetCount(&dwCount));
for (index = 0; index < dwCount; index++)
{
IMMDevice* spDevice;
spEndpoints->Item(index, &spDevice);
GUID subType = {0};
GetJackSubtypeForEndpoint(spDevice, &subType);
if (subType == KSNODETYPE_MICROPHONE_ARRAY)
{
*piDevice = index;
break;
}
}
...// Clean up and return
}
To determine device index,GetMicArrayDeviceIndex:
1.Calls the IMMDeviceCollection::GetCount method to determine the number of devices in the collection.
2.Calls the IMMDeviceCollection::Item method for each capture device to get its IMMDevice interface.
3.For each capture device, passes the IMMDevice interface to the private GetJackSubtypeForEndpoint method to determine the device subtype.
The KSNODETYPE_MICROPHONE_ARRAY subtype corresponds to a microphone array, presumably belonging to the Kinect sensor.
When GetMicArrayDeviceIndex finds this subtype, it returns the associated device index to _tmain.
GetJackSubtypeForEndpoint determines the device’s subtype, as follows:
HRESULT GetJackSubtypeForEndpoint(IMMDevice* pEndpoint,
GUID* pgSubtype)
{
...
IDeviceTopology* spEndpointTopology;
IConnector* spPlug;
IConnector* spJack;
IPart* spJackAsPart;
pEndpoint->Activate(__uuidof(IDeviceTopology), CLSCTX_INPROC_SERVER,
NULL, (void**)&spEndpointTopology);
spEndpointTopology->GetConnector(0, &spPlug);
spPlug->GetConnectedTo(&spJack);
spJack->QueryInterface(__uuidof(IPart), (void**)&spJackAsPart);
hr = spJackAsPart->GetSubType(pgSubtype);
...
}
To determine a capture device’s subtype, you must determine what the capture device is connected to and query that connector for the capture device’s subtype.GetJackSubtypeForEndpoint:
1.Calls the IMMDevice::Activate method to obtain the object’s IDeviceTopology interface.
2.Calls the IDeviceTopology::GetConnector method to get the device’s connector.
3.Calls the IConnector::GetConnectedTo method to determine what the connector from step 2 is connected to.
4.Calls QueryInterface on the object from Step 3 to get its IPart interface.
5.Calls the IPart::GetSubType method to get the capture device’s subtype GUID.
Record the Captured Stream and Determine the Source Direction
After configuring the DMO to capture an audio stream from the Kinect sensor’s microphone array, _tmain calls the private DShowRecord method to record the stream.
Set Up the Data Buffer
The output data is contained in a private CStaticMediaBuffer object that is assigned to the pBuffer member of aDMO_OUTPUT_DATA_BUFFER structure, as follows:
HRESULT DShowRecord(...)
{
...
CStaticMediaBuffer outputBuffer;
DMO_OUTPUT_DATA_BUFFER OutputBufferStruct = {0};
OutputBufferStruct.pBuffer = &outputBuffer;
...
}
For details on the buffer object, see the sample.
Set the Output Format
DShowRecord defines the output format by passing a DMO_MEDIA_TYPE structure with the information to IMediaObject::SetOutputType, as follows:
HRESULT DShowRecord(IMediaObject* pDMO, IPropertyStore* pPS,
const TCHAR* outFile, int iDuration)
{
...
WAVEFORMATEX wfxOut = {WAVE_FORMAT_PCM, 1, 16000, 32000, 2, 16, 0};
DMO_MEDIA_TYPE mt = {0};
hr = MoInitMediaType(&mt, sizeof(WAVEFORMATEX));
mt.majortype = MEDIATYPE_Audio;
mt.subtype = MEDIASUBTYPE_PCM;
mt.lSampleSize = 0;
mt.bFixedSizeSamples = TRUE;
mt.bTemporalCompression = FALSE;
mt.formattype = FORMAT_WaveFormatEx;
memcpy(mt.pbFormat, &wfxOut, sizeof(WAVEFORMATEX));
hr = pDMO->SetOutputType(0, &mt, 0);
MoFreeMediaType(&mt);
...
}
The output is defined by a WAVEFORMATEX structure as follows:
- PCM audio
- 1 channel
- 16,000 samples/second
- 32,000 bytes/second, on average
- A block alignment of 2, which means that the DMO processes 2 bytes of data at a time
- 16 bits/sample
- No extra format information
Allocate Resources and the Output Buffer
DShowRecord next allocates resources, determines the audio frame size, and allocates an output buffer, as follows:
HRESULT DShowRecord(...)
{
...
DWORD cOutputBufLen = 0;
BYTE *pbOutputBuffer = NULL;
...
hr = pDMO->AllocateStreamingResources();
int iFrameSize;
PROPVARIANT pvFrameSize;
PropVariantInit(&pvFrameSize);
pPS->GetValue(MFPKEY_WMAAECMA_FEATR_FRAME_SIZE, &pvFrameSize);
iFrameSize = pvFrameSize.lVal;
PropVariantClear(&pvFrameSize);
cOutputBufLen = wfxOut.nSamplesPerSec * wfxOut.nBlockAlign;
pbOutputBuffer = new BYTE[cOutputBufLen];
cTtlToGo = iDuration * 100;
...
}
IMediaObject::AllocateStreamingResources allocates any resources that the DMO requires.
MicArrayEchoCancellation gets the frame size from the DMO’s property store, by passing the MFPKEY_WMAAECMA_FEATR_FRAME_SIZE property key to the IPropertyStore::GetValuemethod. The output buffer length is the product of the number of samples per second (16,000) and the block alignment value (2). That value is then used to dimension the output buffer. Finally, DShowRecord sets the cTtlToGo value to specify the maximum number of frames to be recorded.
DShowRecord then prepares a file to receive the data. For details, see the sample.
Capture the Audio Stream and Determine Source Direction
Before starting the capture loop, DShowRecord calls QueryInterface on the DMO to get an ISoundSourceLocalizer interface, which exposes the beamforming algorithms that determine the source direction, as follows:
HRESULT DShowRecord(...)
{
...
ISoundSourceLocalizer* pSC = NULL;
...
hr = pDMO->QueryInterface(IID_ISoundSourceLocalizer, (void**)&pSC);
...
}
DShowRecord then starts a capture loop, to record the audio stream, as follows:
HRESULT DShowRecord(...)
{
...
while (1)
{
Sleep(10); //sleep 10ms
if (cTtlToGo--<=0)
break;
do
{
// fill buffer
} while (OutputBufferStruct.dwStatus &
DMO_OUTPUT_DATA_BUFFERF_INCOMPLETE);
if (_kbhit())
{
int ch = _getch();
if (ch == 's' || ch == 'S')
break;
}
}
// Clean up and return
}
The outer while loop captures the audio data one buffer at a time. If the maximum number of frames has been recorded, the capture loop terminates. Otherwise, the inner do-while loop collects the next buffer.
The do-while is essentially a wait loop that cycles until the next buffer is ready. If the buffer is not yet ready, the DMO_OUTPUT_DATA_BUFFER.dwStatus value is set to DMO_OUTPUT_DATA_BUFFERF_INCOMPLETE. After the buffer is ready, the loop checks whether the user has pressed an ‘s’ or ‘S’ key. If so,the outer loop immediately terminates. Otherwise, the outer loop continues to the next buffer.
After the recording process is complete, DShowRecord returns and _tmain performs cleanup and exits.
The do-while loop handles the mechanics of obtaining a filled buffer and determining the source direction, as follows:
do
{
outputBuffer.Init((byte*)pbOutputBuffer, cOutputBufLen, 0);
OutputBufferStruct.dwStatus = 0;
hr = pDMO->ProcessOutput(0, 1, &OutputBufferStruct, &dwStatus);
if (hr == S_FALSE) {
cbProduced = 0;
}
else {
hr = outputBuffer.GetBufferAndLength(NULL, &cbProduced);
}
WriteToFile(hFile, pbOutputBuffer, cbProduced);
totalBytes += cbProduced;
hr = pSC->GetBeam(&dBeamAngle);
double dConf;
hr = pSC->GetPosition(&dAngle, &dConf);
if(SUCCEEDED(hr))
{
if(dConf>0.9)
{
_tprintf(_T("Position: %f\t\tConfidence: %f\t\tBeam Angle = %f\r"),
dAngle, dConf, dBeamAngle);
}
}
} while (OutputBufferStruct.dwStatus &
DMO_OUTPUT_DATA_BUFFERF_INCOMPLETE);
The loop starts by initializing the output buffer object and setting DMO_OUTPUT_DATA_BUFFER.dwStatus to 0. It then calls the DMO’s IMediaObject::ProcessOutput method, which drives processing samples through the pipeline in the following way: