PhraseFind Tips


PhraseFind is one of the more innovative methods available to search lots of footage based on what is spoken, in addition to what might be logged as metadata. This phonetic indexing and search functionality is technology licensed from Nexidia which powers other similar solutions such as SoundBite from Boris. Anyone who has used it will gush over the time-savings and usefulness this brings to workflows. But I bet that many users are not benefiting from the full functionality that PhraseFind offers or how to enter terms correctly to ensure proper results. Unfortunately this information is not part of any Media Composer documentation that I could easily find.

Users can use the following syntax operators to really zero in specific content with extremely high confidence in the results. So… From my modified Nexida documentation on search tips:

A key part of successfully searching audio with Nexidia’s technology is to understand a few basic rules regarding entering search terms. The syntax used here is entered directly in the text entry of the FIND window:

Characters:  A general rule of thumb is to spell out every word in the search query.  This includes:

  • Numbers:  Instead of ‘2008′, type ‘two thousand eight’.  Spell the number using the variation in which it is most likely to be spoken — for example, in an address, ‘495′ is likely to be referred to as either ‘four nine five’ or ‘four ninety five’. 
  • Acronyms:  Separate acronyms that are spoken as a series of letters with spaces.  For example, ‘FBI’ would be entered ‘F B I’ and ‘NCAA’ would most likely be entered ‘N C double A’. 
  • Symbols, Punctuation: Omit all symbols and punctuation such as $, ! and -. 
  • Abbreviations:  Spell out an abbreviation the way it is pronounced.  For example ‘Mr’ should be ‘Mister’ and ‘Dr’ would be ‘Doctor’ (or ‘Drive’).

Quotation Marks:  These are the only non-alphabetic characters that are valid to use in the search box (other than “&” for spanned ranges - see below).  Placing quotation marks around two or more words tells PhraseFind to search for those words together in the sequence they were entered.  For example, entering “President Ford” “United Nations” will launch a search for President Ford as one term and United Nations as the other term.

Spelling:  Because we’re not searching a transcript or other text, correct spelling is not required.  In fact, modifying the spelling of words can actually help improve the results.  If the correct spelling of a word is not pronounced the way in which it is typically spoken, adjust it so that the letters more closely match how it is spoken.  For example, when searching for ‘Barrack Obama’, try ‘Buh rock o bahma‘.

Expression Analysis

This method extracts discrete sequences of adjacent terms that are most likely to have been spoken together, and therefore yield good results for the user:

Multiple Searches:  Given a multi-word search – each individual term, as well as the literal string is independently searched.  An example — President Ronald Reagan – typed into the search box would produce the following search requests:

  • Search (President Ronald Reagan) 
  • Search (President) 
  • Search (Ronald) 
  • Search (Reagan) 

Quotation Marks:  If terms are enclosed in quotation marks, the terms inside the quotation marks are searched together, and the remaining words are each individually searched.  A literal search is not executed when quotation marks are included.  An example – healthcare reform  “Ronald Reagan” – typed into the search  box would produce the following search requests:

  • Search (healthcare) 
  • Search (reform) 
  • Search (Ronald Reagan) 

Span Based Searches

This is very useful when looking for content in a more contextual manner rather than specific instances as noted with the previous method. The FIND interface gives no UI indication for this type of search but is possible if the proper string is entered. For example, if I were cutting a program on the Red Sox winning season from all the broadcast material available, I may want to find footage of Papi and Home Run within a certain time span of each other. The results would indicate that the clip context is probably what I am looking for in this case, versus any time Papi, home, run or “Home Run” are uttered.

The notation is:

  • “Term1 &X Term 2″ … where X is time in seconds.
  • So for my example: “Papi &20 home run” would find those terms within 20 seconds or less of each other

This can also be used to identify 2 or more words within a certain amount of time in a clip.  For example,

  • “Term1 &5 Term2 &10 Term3″

Will locate files with term1 and term 2 within 5 seconds of one another, then within those files, determine if term3 is spoken within 10 seconds of the first 2 terms.  Note this order dependency.

Update: Version 2.0 of PhraseFind re-introduced with Media Composer v8.8 no longer supports span based searches.  

Comments are closed.