You should then play the recorded sound back from the recording device through the input of the sound card and record it onto the computer. The line-in of the sound card is probably the appropriate input. Nowadays, decent quality sound cards are relatively cheap, and so getting reasonable quality is not expensive. When you buy a computer and sound card, you normally get software to enable you to record and manipulate sound. Alternatively, you can buy this relatively cheaply. You will need the software to edit your samples before presentation to the listeners. We used Adobe Audition for this and so the instructions are written for this software, but there are many other pieces of software you could use.
Plug the line out of the recording device into the line in of the sound card.
Open ADOBE AUDITION:
Example of (correct) not overload waveform
Example of overloaded waveform
Ensure that the input data is not overloaded. Overloading is when the input sound is too loud and exceeds the limits of the sound card. In the example shown on the far right the sound volume is too large for the sound card. This is evident by the waveform being clipped, rather than being smooth at the tops. You can also play the sound file and listen to it. If it sounds distorted, this is probably due to overloading. Below are a couple of speech examples in mp3 format:
However, with many noise sources, it is difficult to hear distortion unless it is quite severe, so it is best to inspect the input levels and view the waveform as well. At the bottom of Adobe Audition there is a VU meter showing the level being input. In the image below, it is shown with the levels in green. You should set the maximum level to about -3dB.
To cure overloading, you need to reduce the input level to the sound card. This can usually be done by using the volume control on the computer. Within Adobe Audition this can be done from the following menu choices: "Option", "Windows recording Mixer", "Volume control of Line In".
Extracts from the record sound has to be taken. For instance, the time taken by a washing machine to complete a cycle is too long for a subjective test, and so it is necessary to select sections which represent the characteristic sound of a washing machine. So you must chose which portions of the sound that you wish to reproduce to subjects. Jury testing is a slow process, so it is important that you chose extracts that are distinctly different so that each extract is useful. To take some examples:
In general, the sound samples should be should be at most 30-60 seconds long. Extracts shorter than 10s are very difficult to judge from using the method described later. You should allow about 30s silence at the end of each sample for respondents to make their judgements.
Each sound will be played to each subject once, and so you can now evaluate how long the test will take. In your calculation, allow some time for introducing the tests and auditioning a couple of training sounds for new subjects. It is important that the total duration does not exceed 30 minutes (even better 20 minutes) as tests of longer duration will result in fatigue and people's judgements will gradually get less reliable. If the test does become excessively long, you will have to allow time for subjects to take breaks.
You will find cases where the sound sequence varies with time and is too long to play, for instance the sound of a spinning washing machine takes 15min and the sound isn't constant. In that case, you can choose short representative segments for the spin cycle, with short gaps between. We used 7 segments of the total spin sound to create a period of 30s with a 1s break between two segments. Listen to an example of a shortened sound file.
In most sound processing software, you can select the part of the sound file you want and delete the rest. In the right example using Adobe Audition, you can chose "Edit" "Copy to New" from the top menus after you have highlighted the portion you want. Make sure there are no pops or sounds caused by the sudden starting or stopping of the sound. If this happens, trim the sound or take a slightly longer section to get a clean start and end to the sound.
If you decide to amplify or attenuate a sound file, remember this must be applied to all the sound files of the products. Otherwise the relative volume levels of the instruments will be incorrect. Maintaining the correct relative volume levels is very important because loudness is a dominating feature of perception.
You then need to add to the start of the file the product code, for example "Kettle A". The subjective tests should be run blind, so the product should not be referred to by any distinguishable code. The safest method is to use letters of the alphabet or numbers. This ensures that listener bias ("I always hate the sound of products by company x") doesn't influence results. You might also consider running the tests double blind. This is where the person running the test and analysing the results doesn't know which product is which. In this case another person has to record the sounds and give each one a unique code, before passing to the person running and analysing the results. Only once the results are fully analysed, is the code revealed to work out which washing machine is which. Double blind tests ensures that the test are not biased by the opinion of those running the tests. For example, someone who has spent a long time developing a new product, might inadvertently bias results towards that product.
It is normal to record a short descriptor (such as Kettle A) and append this to the sound file. Listen to an example. To do this in Audition, record the descriptor. Select this recording and select "edit" "copy". Then go to the start of the kettle recording file and select "edit" "paste"
At the end of the file, you might wish to add a silence to enable people to make judgements. Do this by selecting "generate" "silence". We used a 30s period. Alternatively, subjects can pause the playback device (e.g. CD) themselves. Listen to an example.