[{"data":1,"prerenderedAt":379},["ShallowReactive",2],{"project-25099847":3},{"id":4,"name":5,"line":6,"line_source":7,"title":8,"url":9,"domain":10,"maker":11,"posted_at":12,"year":13,"points":14,"comments":15,"rank_in_year":16,"shown_in_year":17,"top":18,"tier":19,"category":20,"subcategory":21,"format":22,"keywords":23,"reception":27,"sentiment":28,"stance_mix":29,"scored_comments":33,"aspects":34,"strip":40,"post_text":43,"what":44,"said":64,"maker_in_thread":157,"shape":167,"grid":171,"attempts":341,"same_week":342,"maker_shows":37,"maker_hidden":371,"reader_version":372,"card":373},"25099847","LibreASR","An On-Premises, Streaming Speech Recognition System","title","Show HN: LibreASR – An On-Premises, Streaming Speech Recognition System","https:\u002F\u002Fgithub.com\u002Ficeychris\u002FLibreASR","github.com","iceychris",1605435020,2020,233,71,218,3028,8,"notable","ai","speech recognition","web_app",[21,24,25,26],"on-premises asr","streaming transcription","rnn-t","curious",0.038,{"positive":30,"neutral":31,"mixed":32,"critical":30},2,18,0,22,{"performance":35,"idea":36,"docs":38,"execution":39},{"positive":32,"critical":30},{"positive":37,"critical":32},1,{"positive":32,"critical":32},{"positive":32,"critical":32},[32,41,41,32,41,41,32,32,41,41,41,32,41,32,32,32,32,32,32,32,32,32,42,41,41,32,41,37,32,32,32,32,41,37,42,41,32,41,41,41],3,5,null,{"problem":45,"audience":43,"pricing":43,"motivation":52,"time_to_build":43,"stack":56,"openness":61,"stage":62,"team_size":63},{"value":46,"source":47,"evidence":48},"Deployable on-premises speech recognition supporting multiple languages","maker",[49],{"id":50,"by":11,"text":51},"25099920","Hey HN! I've been working on this for a while now. While there are other on-premise solutions using older models such as DeepSpeech [0], I haven't found a deployable project…",{"value":53,"source":47,"evidence":54},"No deployable project found supporting multiple languages with RNN-T",[55],{"id":50,"by":11,"text":51},[57],{"value":58,"source":47,"evidence":59},"RNN-T architecture",[60],{"id":50,"by":11,"text":51},"open_source","prototype","unknown",{"praise":65,"objections":74,"requests":91,"questions":113,"alternatives":127},[66],{"value":67,"source":68,"evidence":69},"Good idea to run ESP32-LyraT sample and under-appreciated boards","commenter",[70],{"id":71,"by":72,"text":73},"25104712","kkielhofner","I LOVE that you provided a sample application targeting the ESP32-LyraT! While the ESP8266\u002FESP32 get plenty of love on HN (and elsewhere) I think the ESP ADF (audio development…",[75,84],{"value":76,"source":47,"evidence":77},"Transcription quality is poor compared to cloud services and YouTube captions",[78,81],{"id":79,"by":11,"text":80},"25100007","As I commented above, very poorly. It's still early days.",{"id":82,"by":11,"text":83},"25099938","The upper transcript is YouTube's automatic transcription. Below is the web app transcribing live. And yes, it is actually missing a few words.",{"value":85,"source":68,"evidence":86},"RNN-T is very hard to train and a single GPU is insufficient",[87],{"id":88,"by":89,"text":90},"25106628","nshm","The secret of RNN-T is that it is __extremely__ hard to train, it is very unstable. On a single GPU you'll spend years to train reasonable quality model, moreover a streaming…",[92,99,106],{"value":93,"source":68,"evidence":94},"Add French and other language support",[95],{"id":96,"by":97,"text":98},"25101306","th3h4mm3r","Hi! What should you need to implement other language i.e. Italian or French? I mean: it's a problem due to the less of datas or what? Another question: could you use for example…",{"value":100,"source":68,"evidence":101},"Document CPU and storage requirements",[102],{"id":103,"by":104,"text":105},"25102900","donpdonp","The README could use a section as to what CPU platform and storage requirements are necessary to run this app.",{"value":107,"source":68,"evidence":108},"Report WER per dataset",[109],{"id":110,"by":111,"text":112},"25103212","lunixbochs","Can you post WER per dataset? Bucketing all of the WER together means you can only directly compare to models that are validated on the exact same combination datasets. This…",[114,120],{"value":115,"source":47,"evidence":116},"What datasets are used for training",[117],{"id":118,"by":11,"text":119},"25099954","LibriSpeech, Tatoeba, Common Voice and scraped YouTube videos.",{"value":121,"source":68,"evidence":122},"How does transcription quality compare to cloud offerings",[123],{"id":124,"by":125,"text":126},"25099907","saurik","(Anyone know how the transcription quality compares to the various cloud offerings from AWS, Google, and IBM?)",[128,139,149,153],{"value":129,"source":68,"evidence":130},"Vosk",[131,135],{"id":132,"by":133,"text":134},"25105765","posguy","Mozilla DeepSpeech has streaming audio support as of a few releases ago, the word error rate has also improved. I would recommend looking at Vosk too, it converts speech to text…",{"id":136,"by":137,"text":138},"25103114","woodson","Check out Kaldi. It's a toolkit rather than a ready-to-deploy service but has some solid pretrained models and recipes for training your own. You can use various existing…",{"value":140,"source":68,"evidence":141},"wav2letter",[142,146],{"id":143,"by":144,"text":145},"25101898","whimsicalism","Use wav2letter",{"id":147,"by":144,"text":148},"25102372","> open-source or paid SDK\u002FAPI that I can use to create a group voice chat mobile app with \"live\" transcription? Or something that can plug-in to a system like this? Yes, Google,…",{"value":150,"source":68,"evidence":151},"Mozilla DeepSpeech",[152],{"id":132,"by":133,"text":134},{"value":154,"source":68,"evidence":155},"Kaldi",[156],{"id":136,"by":137,"text":138},{"comments":158,"answered":18,"key":159},12,[160,161,164],{"id":50,"text":51},{"id":162,"text":163},"25101390","Data and compute are the largest hurdles. I only have one GPU and training one model takes 3+ days, so I am limited by that. Also, scraping from YouTube takes time and a lot of…",{"id":165,"text":166},"25117176","Depending on your objective, noisy data might be useful. I'd like LibreASR to also work in noisy environments, so training on data that is noisy should already help a bit with…",{"commenters":168,"max_depth":18,"maker_replies":158,"first_comment_mins":169,"lifespan_hours":170},36,10,272,[172,174,175,176,179,183,186,189,191,193,196,198,200,203,207,208,211,214,217,219,222,225,226,227,230,233,236,239,242,245,248,249,250,252,254,257,259,262,265,267,271,273,276,280,282,284,286,288,291,294,297,299,301,303,306,308,311,314,316,317,322,324,328,329,331,333,336,338,340],[50,32,32,173,11,37],"project\u002Fneutral\u002Fnone",[96,32,41,173,97,32],[162,32,32,173,11,37],[177,32,41,173,178,32],"25101451","jack_pp",[180,32,32,181,182,32],"25101496","other\u002Fneutral\u002Fnone","klysm",[184,32,32,181,185,32],"25103078","Shared404",[187,32,32,181,188,32],"25105785","jerf",[190,32,41,173,144,32],"25101881",[192,32,41,173,97,32],"25101458",[194,32,32,195,97,32],"25101509","other\u002Fneutral\u002Fpricing",[197,32,32,195,144,32],"25101919",[199,32,32,195,97,32],"25105107",[201,32,32,202,144,32],"25105212","commenter\u002Fneutral\u002Fnone",[204,32,41,205,206,32],"25107513","project\u002Fneutral\u002Fperformance","forgingahead",[165,32,32,173,11,37],[209,32,41,173,210,32],"25117415","oulipo",[212,32,32,181,213,32],"25106498","trowngon",[215,32,41,216,144,32],"25101964","maker\u002Fneutral\u002Fidea",[218,32,32,181,137,32],"25103305",[220,37,32,181,221,32],"25100511","guillem_lefait",[223,37,41,173,224,32],"25101206","nsomaru",[143,37,32,202,144,32],[132,37,32,181,133,32],[228,30,32,181,229,32],"25100015","ahoka",[231,30,32,232,11,37],"25100064","maker\u002Fneutral\u002Fnone",[234,30,32,202,235,32],"25100260","Proven",[237,30,32,202,238,32],"25103704","yjftsjthsd-h",[240,41,32,181,241,32],"25100594","fareesh",[243,41,32,181,244,32],"25101684","Y_Y",[246,41,32,181,247,32],"25101837","dcsan",[147,41,32,181,144,32],[136,41,32,181,137,32],[251,41,32,181,144,32],"25103170",[253,41,32,181,137,32],"25103211",[255,41,32,181,256,32],"25110577","deskamess",[258,41,32,181,144,32],"25112346",[260,41,32,181,261,32],"25102918","sbr464",[71,263,42,264,72,32],4,"project\u002Fpositive\u002Fnone",[266,263,32,232,11,37],"25110498",[268,42,41,269,270,32],"25100212","project\u002Fneutral\u002Fidea","blackcat201",[272,42,32,202,11,37],"25100253",[274,42,32,202,275,32],"25100856","bravura",[277,278,41,173,279,32],"25100306",6,"OnlyMortal",[281,278,32,173,11,37],"25100331",[283,278,41,269,279,32],"25100381",[110,285,41,205,111,32],7,[88,18,37,287,89,32],"project\u002Fcritical\u002Fperformance",[289,290,32,181,137,32],"25104630",9,[292,290,32,181,293,32],"25104988","crawfordcomeaux",[295,290,32,181,296,32],"25105175","72deluxe",[298,290,32,181,293,32],"25105249",[300,290,32,181,296,32],"25106913",[302,290,32,181,293,32],"25108081",[304,290,32,181,305,32],"25169020","rthrfrd",[307,290,32,181,293,32],"25222372",[309,169,41,205,310,32],"25099906","pw6hv",[312,169,37,287,313,32],"25099929","jiehong",[315,169,32,173,11,37],"25100223",[82,169,32,205,11,37],[318,319,42,320,321,32],"25105466",11,"project\u002Fpositive\u002Fidea","sagz",[103,158,41,323,104,32],"project\u002Fneutral\u002Fdocs",[325,326,41,327,313,32],"25099939",13,"project\u002Fneutral\u002Fexecution",[118,326,32,327,11,37],[330,326,41,205,270,32],"25100301",[332,326,32,287,11,37],"25100420",[334,326,32,335,111,32],"25103048","commenter\u002Fneutral\u002Fexecution",[337,326,41,269,247,32],"25101801",[124,339,41,205,125,32],14,[79,339,32,287,11,37],[],[343,348,353,358,362,366],{"id":344,"name":345,"points":346,"category":347},"25088683","Stargate (Dwitter 20584)",377,"creative",{"id":349,"name":350,"points":351,"category":352},"25110879","gh-pages-url-shortener",355,"dev_tools",{"id":354,"name":355,"points":356,"category":357},"25139833","Podcast API",223,"data",{"id":359,"name":360,"points":361,"category":347},"25071643","dosylogo",174,{"id":363,"name":364,"points":365,"category":352},"25138872","Scratch.js",169,{"id":367,"name":368,"points":369,"category":370},"25127468","Twelve70",158,"consumer_social",false,"v2",{"id":374,"name":5,"line":6,"ai":371,"cat":20,"year":13,"points":14,"comments":15,"rec":27,"s":375,"top":18,"kw":376,"maker":11,"domain":10,"tier":19,"attempts":32,"strip":377,"url":9,"mix":378,"praise":67,"objection":76,"sub":21},25099847,0.04,[21,24,25,26],"0330330033303000000000533031000031530333",[30,31,32,30],1794036598439]