Forum Replies Created
-
AuthorPosts
-
matukhParticipant
Hi, Halle.
I create replication. Here a shared link for it:https://www.dropbox.com/s/9ipr58wv8azx92u/OpenEarsDistribution.zip?dl=0
I hope it correct created and we decide this problem. Thanks for helping.
Best regards,
VladimirmatukhParticipantHi, Halle.
I change settings following your advices, but I continue to get unexpected results. Can you help me more? And “Once you have confidence in vadThreshold, you can obtain recorded speech as described in the linked post, and start to tune your Rejecto settings (starting from the default settings). If you continue to get unexpected results, you can give me a full replication case as described in the linked post so I can look into whether it’s a settings issue.” where I can look for post about replication.
Best regards, Vladimir.matukhParticipantOk. Thanks for great answer. I will try your advices.
Best, regards Vladimir.matukhParticipantWe send app for testers. And they see all recognized words in textView. For some key words we play audio. After testing they writen to us response about recognition process and about quality. After that we done some changes and sent new version. We done it many times and our team take best result of our testing, but that result does not meet our needs. Did I correctly discribe our testing process?
matukhParticipantok. Our testers read text what contains key word. Recognition response we received fast about 2-5 seconds. Some times it can to take to 10 second. Response we receive after little bit pause. And in response can contain words from our dictionary, but that words are not accrued in the text. Some times when recognition in proccess we can receive response with word “Ruhe” without any speech or noise. Recognition word “Geräusch” look like unreal, after more than 10 spelling we can receive that word.
In initialization of language model generator we set usingVowelsOnly = YES weight = @(1.5). We tested with other parameters but we get worse result(((
Did I answer for your question?matukhParticipantYes, it is correct.
matukhParticipantIt is „Hoch Deutsch“, the „official“ German. Everyone who tests it try to speak as „pure“ German as possible since it is clear that framework will have problems with accents and dialects. Thanks, Halle
matukhParticipantWe put text to Yandex Translate and press Play. Text for example, “Was für ein seltsames Geräusch?” (Key word is geräusch) ( https://translate.yandex.ua/?lang=de-uk&text=Was%20für%20ein%20seltsames%20Geräusch%3F ). And also screenShot http://prntscr.com/eglx5o
matukhParticipantHi, Halle. Thanks for fast response. We testing 2 different ways, by using translator dictation and by people text reading. But result was not well. Some time was detected words from simple sigh.
-
AuthorPosts