Community Wishlist Survey 2020/Archive/Audio lookup of words

From Meta, a Wikimedia project coordination wiki

Audio lookup of words

NoN Outside the scope of Community Tech

  • Problem: Especially with English, it is sometimes the case that someone wants to look up a work that they know how to say but do not know how to spell. Often this is with the goal of finding out how to spell the word. Without already knowing a word's spelling, it is difficult to find words such as phlegm, xylophone, schism or cello.
  • Who would benefit: This will be beneficial to anyone trying to find out how to spell an English word that does not have one obvious spelling.
  • Proposed solution: Use the IPA and/or audio clips associated with a word to generate a hash signature of its pronunciation. Add a feature to the UI to allow users to search by speaking a word. This will record the user speaking the word, generate a hash signature of the recording and then search for words with a matching hash signature. If only one word matches the search show the user that word. Otherwise, first show the user a summary of the words that match.
  • More comments:
  • Phabricator tickets:
  • Proposer: Mgrand (talk) 12:31, 9 November 2019 (UTC)[reply]

Discussion

  • As a French speaker, I can't say it is especially for English. Plenty languages have writing difficulties. Looking for the proper way of writing a word is a regular motivation to use a dictionary for those languages. The proposal Search in a lexicon challenge a similar issue, because a proper search could also be able to deal with IPA. But you are mentioning a search based on audio, and that's something different. Saying something to a microphone to trigger the search could be awesome, and useful not only for Wiktionaries Noé (talk) 17:21, 12 November 2019 (UTC)[reply]
  • This is a great idea! Unfortunately after discussing it as a team, we have concluded this would require long-term engineering efforts that we cannot afford. It is currently not feasible to recognize speech patterns or even to pronounce words from IPA. As such I'm going to archive this proposal as out of scope. Apologies for the disappointment, and thank you for participating in the survey! MusikAnimal (WMF) (talk) 17:11, 15 November 2019 (UTC)[reply]
    Hi MusikAnimal, do you think it could be post next year in the Community Wishlist or is it anyway an idea that is too big? If it is how could we suggest this idea properly, where? Is there's any process to collect this kind of suggestions? Noé (talk) 17:58, 15 November 2019 (UTC)[reply]
    To me it seems like a fairly large machine learning prospect. Google for instance doesn't get pronunciations and speech recognition perfectly either (I think it often relies on context within a full sentence), and they have a lot more resources then us. I think making this happen within our infrastructure would be a lengthy multi-team effort. I would recommend creating a Phabricator task for now, and from there we can hopefully at least put together some technical requirements. I guess use the "Wiktionary" tag. There is also WikiSpeech but this seems to be more about text-to-speech, and not the opposite. As a workaround, if you have a capable phone you could use the built-in speech-to-text in your browser to enter words into the Wiktionary search bar. MusikAnimal (WMF) (talk) 21:56, 15 November 2019 (UTC)[reply]
    Thank you for this feedback. I am aware this proposal is about a feature this is at the edge of actual techs, but I think our projects deserve this kind of very fancy features to attract a new audience. So, I will continue to think about it and then post it in Phabricator. Noé (talk) 07:14, 20 November 2019 (UTC)[reply]