Re: [slim] Double Metaphone based library search
mherger wrote: > I could imagine that FTS and "fuzzy" search could result in a lot of > false positives. Yes, it very likely would. I was envisaging using this as a fallback mechanism - first do a 'normal' text search and if that returns nothing then do a DM search to get back something that I can parse with the FUSEjs library to make sense of the likely falsies. I already use FUSEjs but I need to feed it more data :) philchillbill's Profile: http://forums.slimdevices.com/member.php?userid=68920 View this thread: http://forums.slimdevices.com/showthread.php?t=112531 ___ discuss mailing list discuss@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/discuss
Re: [slim] Double Metaphone based library search
The "Full Text search" plugin (from Logitech) would seem like a good place to put this although I suspect mherger might suggest doing a whole new plugin for it. Not sure yet (lack of investigation). More plugins with inter-dependencies could become a bit complicated to manage. -- Michael ___ discuss mailing list discuss@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/discuss
Re: [slim] Double Metaphone based library search
Just curious if anybody ever experimented with this kind of thing before? Would the increased DB size and time-to-rescan be an issue at all? "Fuzzy search" has been on my list for years... but I haven't done any experimenting yet, tbh. The very first phrase in the wikipedia article about metaphone was somewhat of a bummer: "Metaphone is a phonetic algorithm, published by Lawrence Philips in 1990, for indexing words by their English pronunciation." But it seems DM improved support for other languages. As for the integration... I was wondering whether this should be a different column/index, or whether this could be a feature you enable, and thereafter all search indexes would be using DM. There is a CUSTOMSEARCH column in most tables which could be used by a plugin - and which is put in the fulltext index, too. That said, I could imagine that FTS and "fuzzy" search could result in a lot of false positives. But anyway, it's probably something to explore. -- Michael ___ discuss mailing list discuss@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/discuss
Re: [slim] Double Metaphone based library search
I remember using Soundex back in the early 90s and having just read a bit about Metaphone/Double Metaphone (and even Metaphone 3) this does look like a clever upgrade from it. The "Full Text search" plugin (from Logitech) would seem like a good place to put this although I suspect mherger might suggest doing a whole new plugin for it. Paul Webster http://dabdig.blogspot.com author of \"now playing\" plugins covering radio france (fip etc), kcrw, supla finland, abc australia, cbc/radio-canada and rte ireland Paul Webster's Profile: http://forums.slimdevices.com/member.php?userid=105 View this thread: http://forums.slimdevices.com/showthread.php?t=112531 ___ discuss mailing list discuss@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/discuss
[slim] Double Metaphone based library search
One of the issues I face in developing the Alexa skills for LMS is that I have to search your library -textually- based on the speech-to-text of what Alexa thinks she -heard-. This can be problematic with similar-sounding words like pair/pear, hear/here, knot/not, ate/eight, flour/flower, etc. I can only fuzzy match / post-process the returned values if the searched term is among the search results. Of course a search for a song with -flower- in the title will not return a song with -flour- - end of story. Asking Alexa again will not help because she will hear the same thing every time and return the same search-for text. A potential workaround for this would be to have a plugin that extends the scanning process to add columns to the database that contain computed *Double Metaphone* (DM) values for tags. There is a CPAN lib for DM and all it does is produce a text mapping from a word to something that sounds like the word according to pre-defined rules. Some examples: Flour, flower -> FLR Not, Knot -> NT Hear, Here -> HR Ate, Eight -> AT So the song title 'True Colors' would become TRKL, but so would Tru Colors, True Colours and even Trew Cullers. This means that I can match based on *sound* of the title rather than spelling and therefore more easily find stuff in your library. With the added benefit that this would work in the GUI interface too and help you find stuff with misspelled tags. Just curious if anybody ever experimented with this kind of thing before? Would the increased DB size and time-to-rescan be an issue at all? philchillbill's Profile: http://forums.slimdevices.com/member.php?userid=68920 View this thread: http://forums.slimdevices.com/showthread.php?t=112531 ___ discuss mailing list discuss@lists.slimdevices.com http://lists.slimdevices.com/mailman/listinfo/discuss