All sound effects database systems extract and store file-related attributes. These typically include file name, creation date, and other items easily found through the native file system, or within the header of the sound file being added; for example, sample rate, number of channels, sample width, sample format, file format, and duration. These extractions take trivial amounts of CPU time. Additionally, all systems can store text descriptions (e.g., keywords, comments, etc.), generally entered by the user after the sounds are added to the database. SoundFisher performs all the extractions mentioned above, but it also analyzes the sound samples directly, computing perceptually salient features of the sound itself. This is a CPU-intensive operation. The result is a set of statistical approximations, referred to as a "feature vector", which capture the spectral and amplitude characteristics of the sound itself.Benchmarks indicate that on average, SoundFisher analyzes and adds sound to the database at approximately 30 MB per minute on an Apple Macintosh G3 running at 300 MHz. This is roughly equivalent to 3-times real time, if you are adding CD-quality sound files.(Note that the default segmentation preference ("None") forces the add operation to analyze a maximum of the first 15 seconds of each sound, which means that each sound will take no longer than roughly 5-to-7 seconds to add to the database on the benchmarked G3 system, above. For further information about segmentation and it's affect on analysis timing, see "What is automatic segmentation..." below.)

Automatic segmentation is the virtual subdividing of a sound file as it is analyzed and added to the database. For each subdivision (i.e., "segment") of the input sound, SoundFisher computes a separate feature vector containing a numeric representation of perceptually salient qualities of the sound. (Note that the input sound file itself is not altered in any way during the segmentation and analysis process.) You choose the type of segmentation you want by using the Preferences->Segmentation Properties... menu item. You can change segmentation type any time. Each time you invoke the Add Sounds process the specified input sounds will be segmented and analyzed according to the current segmentation type. A SoundFisher database can contain sounds that were added with a variety  of segmentation types, however only one segmentation type can be applied to a given sound. (This is mainly due to a limitation in the current product, which bars you from adding a sound to a database multiple times.) Currently, SoundFisher offers three types of segmentation: "None", "Silence" and "Fixed".The default, None, treats the input sound file as a single segment. (Therefore, the sound is really not segmented at all.) The analysis and resulting feature vector is computed on a maximum of the first 15 seconds of the input sound file, and this is considered to be the sound's representative (and only) segment.  This setting is appropriate for most SoundFisher databases, which typically contain short sound-effects (1 to 15 seconds in duration) or longer but homogenous sounds. Examples of the former include glass breaking, gun shots, car starts, brief Foley sounds, etc. Examples of the latter include indoor or outdoor ambiences, vehicle "pass-bys", rain on pavement, etc.Alternatively, the Silence segmentation type will cause the Add Sounds process to analyze the entire input sound file. Whenever SoundFisher encounters a block of "silence" (e.g., leader) within the sound file, it effectively pauses the analysis and generates a feature vector which describes the sound segment that occurred between the end of the previous and the beginning of the current leader. Silence is specified as a noise threshold, typically -60 dB. This setting is appropriate, for example, when adding multi-take recordings, where leader (actual silence) separates each take, or when adding field recordings where the the sounds of interest are separated by stretches sound whose amplitude falls below the specified noise threshold. Lastly, the Fixed segmentation type will cause the Add Sounds process to (virtually) chop the input sound file into a sequence of fixed-duration segments and to produce an analysis feature vector for each one. (You specify the segment duration.) Fixed segmentation can be useful when adding long heterogeneous sounds that are not broken up by leader or very low amplitude background sound. For a detailed explanation of each segmentation type, see "Segmentation of long files" in the User Manual. For details about the segmentation properties associated with each type, see "Segmentation Properties..." in the "Setting preferences" section of the User Manual.

When you add a sound to a SoundFisher database, the application first divides the sounds into one or more segments, then analyzes each segment, producing a very compact description of perceptually relevant aspects of that segment. This description is called a feature vector. SoundFisher's analysis is a powerful audio data reduction technique, of sorts. A SoundFisher feature vector is a 144-byte structure. It contains statistical data (primarily, the mean and standard deviation)  for numerous acoustic and perceptual attributes. Simply put, these attributes describe the pitch, brightness, bandwidth and loudness characteristics of the sound.
All SoundFisher feature vectors contain the same amount of data; they are equal sized, regardless of the duration of the sound segment that a feature vector describes. The feature vector consists of the computed mean and standard deviation for each sound attribute. Therefore, the longer and more heterogeneous the segment, the less accurately the feature vector will capture the segment's characteristics. Put another way, the details of very long and heterogeneous sound segments will become overly smoothed or washed out by the analysis.  For this reason, it's important for you to be at least somewhat familiar with the sound files you are adding to SoundFisher, so that you can set the segmentation type properly. In general, the analysis which produces the feature vector does a good job of representing most segments whose duration is about 15 seconds or shorter, even if the sound within that segment varies widely. For more information about which segmentation type to apply when adding sounds, see "What is automatic segmentation and what is proper setting for segmentation type?" above.

A thumbnail is generally a low fidelity, possibly abbreviated, version of the original sound file added. When adding sounds from a CD-ROM or other removable media, SoundFisher will automatically generate a thumbnail version of each sound and store it locally on the the same hard drive that contains the database. This permits SoundFisher to playback a rendition of the original sounds, even if the originals are now offline. SoundFisher can also create a thumbnail of any sound after it has been added to the database.

SoundFisher's analysis computes statistical representations of the sound. Therefore, by definition, it is an approximation technique. When SoundFisher performs a sounds-like query, it computes the distance between each pair of properties found in two feature vectors, arriving at an approximate (statistical) measure of similarity. In theory, the smaller the overall distance between two feature vectors, the greater the perceived similarity between the sounds compared. Of course, given the approximate nature of the analysis and the comparison, the results are not always what you would expect (or like). Sounds-like queries can produce some puzzling results. But, more often than not, assuming your database contains a rich and plentiful mixture of sounds, the results can be surprisingly accurate. (We generally find that 60 to 80 percent of the top ten sounds resulting from a query do indeed have definite and perceptible similarities.) Producing good sounds-like query results is dependent on many variables, not the least of which is the number and range of sounds that you've added to the database. The best results start occur when you have a database consisting of at least 500 disparate sounds and when there are at least several perceptually similar variations of each type of sound within the collection.The worst results occur when the database contains too few sounds, or where all of the sounds are perceptually quite similar (e.g., all of sounds are interior recordings of train stations or of traffic on city streets), or where none (or very few of the sounds) bear any resemblance to each other.Sometimes you can get better results by selecting more than one example sound. Providing the query with multiple perceptually similar example sounds gives the search engine more specific information about the kind of sounds you are seeking. Note however that providing more examples does not always produce better search results. Searching with SoundFisher is somewhat of an art: the more you do it, the better you can get. By experimenting and playing with SoundFisher, you will learn what kinds of sounds-like queries work best, given your particular archive of sounds.

At present, SoundFisher cannot automatically import your existing databases, however we can work with you and customize a version of SF that does import the text fields and file paths from your current database.

Currently, there is no direct method of weighting the importance of specific features, however, SoundFisher tries to deduce weightings by looking at the particular values of each feature. Suppose you select an example sound that has a definite pitch. SoundFisher will note this fact and weight pitch comparison heavily. This search should result in other sounds that have a similar pitch content. Or, suppose you select two perceptually similar example sounds for your query. SoundFisher will again try to determine which features make these two sounds perceptually similar by looking for strong correlations. So, if the differences between the sounds' brightness characteristics (means and standard deviations) very small, but the differences between the other characteristics of the two sounds are large, SoundFisher's query engine will automatically weight brightness comparisons. So, to some extent, you can help SoundFisher weight the importance of specific features by your choice of example sounds for the query. Sometimes you can get better search results by combining sounds-like with other query fields. For example, if you are looking for sounds that are perceptually similar to the example you've selected and are less than 3 seconds in duration, you can add a query element for the specification of duration. To build a query that combines multiple fields, use theQuery->Add Field menu item.

No. The current version of SoundFisher only supports ANDing query fields.

The example sound (or sounds) you select for a sounds-like query guides the comparison engine during a search. When you find a sound or combination of sounds that yield good search results, you may want SoundFisher to remember the example for future searches. This will save you from having to manually find and select those example sounds the next time you want to run that search. Let's suppose you manually select one or more sounds of glass shattering and use them as the example for a sounds-like search, and the search produces some good results. After the search completes, click the "go to previous database view" button (i.e., the left-pointing arrow button) in the browser control panel. This will take you back to the view from which you ran the search, and the example sounds will still be selected. Next, invoke the Classes->Train Sound Class... menu item. This will open a dialog that lets you save the example as a "class" (i.e., prototype), for subsequent searches, whether you perform the search again in the current SoundFisher session or in some future session. This dialog lets you assign a name to the class. Suppose you name it "breaking glass". The next time you set up a sounds-like query, press and hold the mouse button on the value field of the query (i.e., the right-most component of the Sound query). A pop-up menu will display the default menu item, "selection", as well as any named sound classes that you have built. If you choose the class you created above, "breaking glass", that trained class will be used as the example that guides the subsequent search.One of the useful side effects of creating and using sound classes is that they remain available for queries, even if you remove the sounds that were used to train the class. Future versions of SoundFisher may come with pre-built sound classes that can be used to find many kinds of sounds without requiring you to manually locate and select the example sounds.

When you add sounds to a SoundFisher database, the application creates categories that match the names of the file system folders containing those sounds. So, if you add sounds that reside on your hard drive in a folder called "My Sounds", SoundFisher will create a category in the database named "My Sounds" (assuming it does not already exist) and add the sound analysis and other information about the sounds to that category.  So, after initially adding sounds to a SoundFisher database, the categories (as displayed in the Category View) will look identical to the folder hierarchy of the native file system.

Like competing products, SoundFisher does not automatically update the database if you rename, move or remove a sound or folder from your file system. Since SoundFisher only adds meta data to a database -- as opposed to adding the actual sound samples -- it isn't bothered by the fact that a sound file was renamed, moved or removed from the actual file system. You will still be able to run queries, train classes, etc. However, if SoundFisher tries to play a sound that was renamed, moved or removed after being added to the database, it will display a dialog indicating that the target sound cannot be found.This dialog gives you the options of showing SoundFisher where the original input file is now located on the file system or, in the case of a rename, what its new name is. If the original file was removed from the file system, you may want to select the dialog option that removes the sound meta data from the SoundFisher database. Note exceptions: if SoundFisher stored a thumbnail of the original input sound, either when it was added to the database -- as is the default when adding sounds from a removable medium --  or later, it will simply play the thumbnail without complaining that it cannot find the original. You can ask SoundFisher to "resynchronize" with the file system at any time. The File->Synchronize... menu item will check the database for any unresolved links to sound files and give you the options described above.

To create a new sound category in the SoundFisher database, first select Category View in the view-type pop-up menu. Then, invoke the Edit->New Category... menu item. This will display a New Category dialog where you enter the new category's name. Press the "Yes" button in the New Category dialog. The category you specified will be created as a sub-category of whatever category was selected prior to invoking the Edit->New Category... menu item. If no category was selected, the new category is created under the root (i.e., "top") category. To move an entire category, including the sounds contained within that category, simply drag the category (represented by a folder icon in the Category View) and drop it on the desired category. The dragged category will now become a sub-category of the target category (i.e., the category onto which the original was dropped). Note that depending on how many sounds the dragged category contains, moving it can be a time consuming operation. (We are working on speeding up this process.) To move one or more sounds from their current category to a different category, make sure you are in Category View then simply select, drag and drop the sounds to the desired category (i.e., folder icon). Again, if you are attempting to move many sounds, this can be a time consuming operation.

SoundFisher can export (send) selected sounds to any application that responds when it receives sound file. Before you can export sounds, you must "register" the desired applications with SoundFisher. This is done through the Preferences->Editors/Players... menu item, which will open a dialog. Click this dialog's Browse button and navigate your file system to locate an application you want to use with SoundFisher. Repeat this "Browse" process to register other applications. (You'll probably want to register your sound editing applications and sound players.) When you're done, close the dialog. Now, select any sound in SoundFisher. Notice that the menu item, File-> Export Selected Items to,  allows you to choose any of the applications you registered. Choose one. In a few seconds, SoundFisher will cause the application to launch and open the selected sound or sounds. Please see User Manual, section "Exporting sounds to external applications", for details about sending sounds to particular applications (e.g., Digidesign's ProTools).

Recent Mac releases of ProTools (5.1 LE, for example) include a new option that prohibits other applications from sharing the Digidesign hardware while ProTools is running. To enable sharing, turn off the Active in Background option found in the ProTools Operations menu.

Not yet, but we are working on this with some sound effect distributors. If you are the licensor or distributor of commercial sound effects and would like to bundle SoundFisher with them, please contact us at soundfisher@musclefish.com