Installation_Instructions

• Supported Platforms
  • Building From Source
  • OSX
  • Windows
  • Linux
• General Information
  • openStreamManyEars
  • EqualGains
  • LocalizeBeam
  • SeparGSS
  • SourceTrack
  • ToString
  • QtSendString
  • Constant (CONDITION)
  • SaveWavWin32
  • UDPOSCOut
    • Important Notes
  • Microphones Configuration
    • Important Notes
  • Audioviewer

Supported Platforms

• Make sure you have installed the FlowDesigner and ManyEars packages before starting. Binaries for OSX, Windows & Linux can be found on the SourceForge download section L
  • FlowDesigner (Base data-flow system)
  • ManyEars (ManyEars plugin)

If binaries are not available for your installation, the following section contains information on how to build from source.

Building From Source

Please consult the [Downloads] section to get the code from SVN or the latest tarball. Instructions for supported operating systems are next.

OSX

Windows

**Notes:**
* We are assuming here that you are using Dev-Cpp to get pre-compiled (binary) packages from community devPaks.
* We are also assuming that FlowDesigner is installed in C:\Program Files\FlowDesigner

• 1) Follow the instructions to compile FlowDesigner first.
• 2) Install the FFTW3 package from Dev-Cpp

• 3) Start a terminal, you are now ready to generate the make files and type in the ManyEars root directory :

  cmake -G "MinGW Makefiles"

• 4) Compile & Install ManyEars:

  make install
  (The default directory is C:\Program Files\FlowDesigner\lib\flowdesigner\toolbox\MANYEARS. This can be changed in the root CMakeList.txt)

• 5) Run flowdesigner.exe from C:\Proram Files\FlowDesigner\bin. The ManyEars toolbox should be available in the TreeView.

Linux

• 1) Install FFTW3

  apt-get install fftw3-dev

• 2) Generate Makefiles

  cmake .

• 3) Compile & Install

  make;make install

General Information

The main processing loop contains the following elements :

[[img src=ProcessingLoop.png width=640]]

openStreamManyEars

• Using RTAudio to get audio stream.
• We are sampling at 48kHz.
• Frames are 1024 samples wide, with 50% overlap.
• Each input is interlaced (8) and sampled with 16bits little endian.

EqualGains

• 8 gains are defined for each microphone input. This is useful for fine tuning gains for slightly different microphones or hardware sampling.
• Gains are applied for each sample.

LocalizeBeam

• Localization algorithm with the steered beam former. Please refer to the publications for more information.

SeparGSS

• Geometric sound source separation. Please refer to the publications for more information.

SourceTrack

SoureTrack will output a Vector of SourceInfo, which is the data structure that contains active sound sources in FlowDesigner. Most useful informations are :

1. float x[3] Contains x,y,z coordinates of the detected source. This is related to the origin of the microphone array positions.
2. float strength Strength of the source.
3. int source_id Id of the source (unique).

ToString

• Transform the Vector<ObjectRef> output of SourceTrack containing source informations into a FlowDesigner serialized string.

QtSendString

• Will send the string to a TCP socket on port 30000. This is the port of the audioviewer application for displaying live sources.

Constant (CONDITION)

• This is the loop condition. The constant is set to "TRUE" for infinite looping.

SaveWavWin32

• Separated streams will be saved in WAV format.

UDPOSCOut

Open Sound Control output using UDP sockets.

• OSC format for ManyEars is the following :

  "/manyears\0\0\0,iffffff\0\0\0\0"
  + source_id (4 bytes)
  + x (4 bytes)
  + y (4 bytes)
  + z (4 bytes)
  + strength (4 bytes)
  + theta (4 bytes)
  + phi (4 bytes)

The resulting code is :

int id = info->source_id;
float x = info->x[0];
float y = info->x[1];
float z = info->x[2];
float strength = info->strength;
float phi = -180.0 * atan2(info->x[2], info->x[1]) / M_PI;;
float theta = 180.0 * atan2(info->x[1],info->x[0]) / M_PI;

stringstream outputString;

char header[12] = {'/','m','a','n','y','e','a','r','s', '\0', '\0','\0'};
char tags[12] = {',','i','f','f','f','f','f','f','\0','\0','\0','\0'};

outputString.write(header,12);
outputString.write(tags,12);

BinIO::write<int>(outputString,&id,1);
BinIO::write<float>(outputString,&x,1);
BinIO::write<float>(outputString,&y,1);
BinIO::write<float>(outputString,&z,1);
BinIO::write<float>(outputString,&strength,1);
BinIO::write<float>(outputString,&theta,1);
BinIO::write<float>(outputString,&phi,1);


//write to the socket
socket.writeDatagram(outputString.str().c_str(),outputString.str().size(),
QHostAddress(hostname.c_str()),portnumber);

Important Notes

• Phi and Theta can be calculated from x,y,z. We have decided to transmit them anyway for ease of use.
  • float phi = -180.0 * atan2(info->x[2], info->x[1]) / M_PI;;
  • float theta = 180.0 * atan2(info->x[1],info->x[0]) / M_PI;
• Data must be aligned on 32 bits with OSC.
• tags[12] = {',','i','f','f','f','f','f','f','\0','\0','\0','\0'}; describes the format of the OSC packet. Binary data is written using big endian (network order).
• We send N separated packets, one for each of the active source.
• We should consider writing a generic OSC FlowDesigner Node.

Microphones Configuration

Default microphone configuration is defined in file cube_mic_pos.mat :

[[img src=Cube.jpg]]

• Microphone #1 x=-18cm y=16cm z=-15.5cm
• Microphone #2 x=-18cm y=16cm z=15.5cm
• Microphone #3 x=18cm y=16cm z=-15.5cm
• Microphone #4 x=18cm y=16cm z=15.5cm
• Microphone #5 x=-18cm y=-16cm z=-15.5cm
• Microphone #6 x=-18cm y=-16cm z=15.5cm
• Microphone #7 x=18cm y=-16cm z=-15.5cm
• Microphone #8 x=18cm y=-16cm z=15.5cm

Important Notes

1. Microphones must be ** connected in the correct order ** in the sound card inputs. Microphone #1 goes into input #1, Microphone #2 goes into input #2, etc.
2. Microphones positions can be changed to fit your setup. It is mandatory that you configure the microphone positions otherwise the algorithm won't work.
3. Sampling time for each audio input must be synchronized by hardware.

Audioviewer