GuineaPig - Overview guinea pig n 1: a small stout-bodied short-eared nearly tailless domesticated rodent (Cavia cobaya) often kept as a pet and widely used in biological research 2: a subject of scientific research, experimentation, or testing - Webster

The goal of the GuineaPig project is to build a system that makes easier to create psychoacoustical listening test by taking care of some tedious routine tasks like creation of playlists. System will also automate the running of these tests.

Traditional procedure when making listening tests is such that the tester edits and records the whole test on a DAT-tape that is then played to the test subject. The test subjects listen to the test and mark their answers on paper. The tester then has to enter the answers manually to a computer to analyze them. GuineaPig eliminates the need to edit the whole test into a tape beforehand and removes the manual entering of the results. Also the test doesn't need to be the same for all subjects. The order, in which the individual test items consisting of sound samples are presented to the subject, can be very easily arranged to be different for each session.

The first phase is the test creation where the tester selects the test type, used sound samples and other test parameters. As the result configuration files are generated that contain the parameters needed in the test.

The test is then run for each test subject and results are saved. After the tests have been performed the results can be studied, processed and analyzed with some statistical programs.

Test types

• A/B: Test in which the test subject chooses one out of two samples played to him/her.
• A/B/X: Test in which the sample X is the same sample as A or B. Subject chooses which one sample X is.
• A/B/C: Test in which A is the reference sample and the subject chooses how much samples B and C (one of which is the same sample as A) resemble the reference sample A. (Actually this test is referred as a Ref/A/B-test).
• A/B Scale: An A/B test in which subject gives an answer for both samples on a scale specified by the test creator.
• A/B Scale, Fixed Reference: A test in which subject gives an answer how sample compares to a reference.
• A/B Scale, Hidden Reference: A test in which subject gives an answer for both samples on a scale specified by the test creator. One of the samples A or B is the hidden reference.
• Single Stimulus: A Test in which a single stimulus signal is played and then graded.
• TAFC (Two alternatives forced choice): Test in which two samples are played altering some parameter, until subject can no longer hear difference between the samples.

Testing features

• Multiple questions: All tests can include any number of questions. GP contains ready-made question components for giving grades, multiple-choices, and rank-order. Also additional question components can be added.
• Free or fixed sequence: The sample sequence can be fixed (freely configurable with multiple samples and pauses between samples) or free where the subject selects in which order samples are played.
• Time limit for answering: A timeout can be set to limit the time the subject has for giving the answers for a test item. Without a timeout the subject can have as much time as he needs to decide the answers. With a timeout, the test goes to the next item when the time ends.
• Parallel switching: Normally when the subject switches to another sample, the currently playing sample stops and the another starts from beginning. In parallel mode the switching to another sample is done by cross-fading to the other sample and does not just jump to the start to the beginning to the sample. Useful in test where long samples are compared. Parallel switching is only usable for free sequence tests.
• Playlists are used to define the order of the test items that are presented. Different playlists may be defined for different sessions.

Audio output features

The Sound Player uses the SGI Audio Library and the SGI Audio File Library, therefore all supported devices and sound file formats supported by them are available in GP. The audio devices includes the normal analog stereo outputs on most systems as well as the default or additional digital outputs up to eight digital channels on one device (ADAT). The supported audio file formats include: AIFF/AIFF-C, Next/Sun, Wave, MPEG1, raw data, and many more.

Some features of the sound player:

• Multichannel: Number of output channels is limited only by output devices of the machine. The sound player can use multiple audio devices in parallel, emulating wider output port than a single device. For example, when three ADAT cards are used, 24-channel output is possible. Multiple devices are syncronized automatically.
• Virtual players allow partitioning the output of GP into smaller sections. For example, single ADAT output could be used to implement four stereo players.
• Many audio file formats: All audio file formats supported by SGI's audio file library including: AIFF/AIFC, WAV, AU, MPEG-1 layers I/II and others with multiple sample rates and sample widths.
• Sampling rates: depending on the audio device, at least the most common rates from 8kHz to 48kHz are supported (device may limit the possible choices. For example, digital outputs generally support only 32/44.1/48 kHz).
• 24 bit audio: SGI audio libraries can support up to 24 bits per sample. GP's sound player uses floating point calculation for rendering audio output.
• Analog and digital outputs: Most SGI workstations have analog outputs. Some also have digital outputs by default (Octane, some Origin and Onyx2 servers) or with an digital audio option (on O2 with the PCI Professional Audio Option). Digital output interfaces include ADAT/SPDIF/AES3, optical/RCA/BNC connectors. Analog output generally are 16bit, digital upto 24 bit (ADAT/AES3).
• Sample mixing: multiple samples can be mixed to the output with each with its own volume levels and faders.
• Sample synchronization: samples can be synchronized with sample frame accuracy.
• Faders: each sample has its own fader that does real fades using linear or dB linear fading. Also supports cross-fading between samples.
• Volumes: each sample sample has its on volume controls (with fader, see above). Also there is a calibration level for each sample that can be used to calibrate a set of samples used in a test. Player's output also have a calibration or master volume level control than can be used for setting the MCL level of a test session of general master output level.
• Java module for controlling sound player: the sound player is a separate program. A easy object-oriented java module is provided for controlling the player.

Requirements

Most of the system is written in Java. Java 1.1 is required and it is freely available for download from SGI. In future, Java 1.2 (Java2) will be used.

Additional information

Jussi Hynninen, A software-based system for listening tests, 2001

The GuineaPig web-page is located at:

www.acoustics.hut.fi/projects/GuineaPig2/

The page contains the AES 106th GuineaPig paper and presentation slides presented at the AES 106th Convention, May 8-11 1999, Munich, Germany:

• GuineaPig paper in AES 106th: [ gzipped PS (674K) / PDF (1103K) ]
  
• AES Presentation slides on 25/May/1999: [ gzipped PS (256K) / PDF (352K) ]
  Slides shown at the paper presentation at 106th AES Convention. Includes extra slides that were excluded from the presentation to fit to the time allowed.