Skip to content

Latest commit

 

History

History
171 lines (119 loc) · 6.67 KB

PROTOCOL.md

File metadata and controls

171 lines (119 loc) · 6.67 KB
Raw

Experimental Protocol

General

Room EPFL, ELA 006 Anechoic Chamber

Date 2017 / 09 / 26

Time Start 12:19

Time End 19:46

Team Robin Scheibler, Hanjie Pan

Description of the Experiment

The microphone array pyramic with 48 MEMS microphones was placed on K&B turntable and rotated 360 degrees in increments of 2 degrees for a total of 180 distinct angles. Three loudspeakers were placed at fixed locations, all forming the same angle with the array, but with 3 different heights.

setup

A total of 8 samples were played by each loudspeaker for each angle:

  • Linear sweep, samples/sweep_lin.wav, 48 kHz
  • Exponential sweep, samples/sweep_exp.wav, 48 kHz
  • White noise, samples/noise.wav, 48 kHz
  • 3 female speech samples, samples/fq_sampleX.wav, X=0,1,3, 16kHz
  • 2 male speech samples, samples/fq_sampleX.wav, X=2,4, 16kHz

The sweeps were generated by running

python ./code/gen_sweeps.py

The random noise sample was generated using the numpy.random.randn function.

The speech samples were picked from the TIMIT corpus as follows

  • fq_sample0 : FAKS0 / SA1 (female)
  • fq_sample1 : FCMR0 / SA2 (female)
  • fq_sample2 : MDLD0 / SX13 (male)
  • fq_sample3 : FMGD0 / SX34 (female)
  • fq_sample4 : MJLN0 / SX99 (male)

All the samples were resampled at 48 kHz if necessary and stitched into one file (all_samples.wav) before playback. The playback was done at 48 kHz.

anechoic_room

For each angle, the array was rotated by the appropriate amount programatically using the computer interface from the turntable. Then, each loudspeaker played all_samples.wav while the Pyramic array was recording.

The recorded file was saved as

recordings/pyramic_spkrX_all_samples_Y.wav

where X is the index of the speaker with the following mapping:

X=0 => Middle loudspeaker
X=1 => Low loudspeaker
X=2 => High loudspeaker

and Y is the counter-clockwise rotation angle in degrees of the array. The corresponding relative rotation of the speaker around the array is 360 - Y.

Experimental Data and Equipment

The file protocol.json is machine readable and contains both the environmental conditions (temperature and humidity), the material used, and the geometry of the experiment (the relative locations of microphones and loudspeakers).

Format description

We use machine readable JSON format with the following fields.

  • conditions : Environmental conditions during the experiment
    • start/end : start and end time of experiment
  • equipment : Equipment used during the experiment
    • interface : The audio interface used for the playback of sound samples
    • speakers : The brand and model of loudspeakers
    • turntable : The brand and model of the turntable on which the microphone array lies
    • microphones : The microphone array used for the experiment
  • geometry : The geometry and locations of objects during the experiment
    • microphones
      • locations : A list of triplets describing the locations of the microphones relative to the center of the array
      • height : The height of the lowest microphone in the array with respect to the floor
      • comment : A short description of the other fields
    • speakers

      • locations : Location of the loudspeakers with respect to the array

        • middle/low/high : Names for the three loudspeakers locations
          • distance : Horizontal distance between the front of the loudspeaker and the center of the array
          • height : Vertical distance between the bottom of the wooffer baffle and floor

      • baffles_corrections : The vertical distance between the bottom of the woofer and the center of the woofer and twitter baffles

      • comment : A short description of the above fields

Code Automating the Experiment

The scripts used for automating the experiment are stored in the code folder. Each script contains a short description of its purpose.

  • gen_sweeps.py generates a linear and an exponential sweep, as well as their time windowed versions
  • run_experiment.py runs the full experiment
  • segment.py segments the recorded files into separated files corresponding to each original sample. It resamples the recorded speech samples at 16 kHz.
  • stitch_samples.py resample and stitch together all the samples and creates all_samples.wav
Dependencies

Python 3.5.3 was used to run the experiment. The following three standard modules are required and can be installed via pip.

In addition, these two custom modules should be downloaded. Edit EASYDSP_PATH and TURNTABLE_PATH in run_experiment.py to indicate their locations.

  • Turntable Driver Commit: 27446ae17888e902a51f4867b59bd20064135fc7
  • Easy DSP Commit: 16b4e807a8fe1d82caa81baf6345446fd6e655cb

Identify Turntable device name

>>> import visa
>>> rm = visa.ResourceManager()
>>> rm.list_resources()
('ASRL1::INSTR', 'ASRL2::INSTR', 'ASRL3::INSTR', 'ASRL4::INSTR', 'GPIB0::12::INSTR')

Run the experiment

The data was collected by running

python code/run_experiment.py -a pyramic -r 0:360:2 -o 2 -t ASRL4::INSTR 192.168.0.101

where -o 2 indicated the number of the playback device, -t ASRL4::INSTR the address of the turntable, and 192.168.0.101 the IP address of the Pyramic array.

About Synchronization of Playback and Recordings

There is no synchronization between playback and recording. The reason is that the Pyramic array used for recordings currently lacks the capability of controlling a loudspeaker synchronously with the acquisition. As a consequence, the absolute time of the recorded samples is unknown. This means, for example, that the time-of-flight (TOF) cannot be obtained from the samples.

Notes

  • Bugs with the USB interface crashed the computer several times and the measurements were thus taken by batch. The batches are the following.

    • Batch 1: Angles 0 to 14
    • Batch 2: 16 to 180
    • Batch 3: 182 to 190
    • Batch 4: 192 to 206
    • Batch 5: 208 to 216
    • Batch 6: 218 to 358

  • During quality control a glitch in recording at angle 310 with spkr 0 was found and the measurement was repeated on the next morning, but before the setup was moved. When the measurement was repeated, the temperature was 21.6C and the humidity 48.9%. The glitched sample is preserved in the 'discarded' directory."