[{"data":1,"prerenderedAt":486},["ShallowReactive",2],{"project-47143755":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":34,"aspects":35,"strip":43,"post_text":45,"what":46,"said":62,"maker_in_thread":216,"shape":218,"grid":221,"attempts":449,"same_week":450,"maker_shows":32,"maker_hidden":479,"reader_version":480,"card":481},"47143755","Moonshine Open-Weights STT models","higher accuracy than WhisperLargev3","title","Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3","https:\u002F\u002Fgithub.com\u002Fmoonshine-ai\u002Fmoonshine","github.com","petewarden",1771970047,2026,316,81,135,4366,4,"legend","ai","speech to text","library",[21,24,25,26],"streaming transcription","open weights","on-device speech recognition","mixed",0.12,{"positive":30,"neutral":31,"mixed":32,"critical":33},20,22,1,8,51,{"performance":36,"idea":39,"docs":41,"design":42},{"positive":37,"critical":38},15,3,{"positive":18,"critical":40},0,{"positive":40,"critical":38},{"positive":40,"critical":40},[32,40,40,32,40,40,40,40,44,44,38,40,44,44,44,44,40,44,44,38,32,38,32,44,40,44,44,38,44,38,44,38,44,38,32,32,32,38,40,44],5,"I wanted to share our new speech to text model, and the library to use them effectively. We're a small startup (six people, sub-$100k monthly GPU budget) so I'm proud of the work the team has done to create streaming STT models with lower word-error rates than OpenAI's largest Whisper model. Admittedly Large v3 is a couple of years old, but we're near the top the HF OpenASR leaderboard, even up against Nvidia's Parakeet family. Anyway, I'd love to get feedback on the models and software, and hear about what people might build with it.",{"problem":47,"audience":53,"pricing":53,"motivation":54,"time_to_build":53,"stack":58,"openness":59,"stage":60,"team_size":61},{"value":48,"source":49,"evidence":50},"Accurate speech-to-text that can stream locally with low word-error rates.","maker",[51],{"id":52},"post",null,{"value":55,"source":49,"evidence":56},"Create streaming STT models with lower word-error rates than Whisper Large v3.",[57],{"id":52},[],"open_source","launched","team",{"praise":63,"objections":92,"requests":126,"questions":164,"alternatives":195},[64,84],{"value":65,"source":66,"evidence":67},"Streaming latency is very fast, including on CPU and M1 hardware.","commenter",[68,72,76,80],{"id":69,"by":70,"text":71},"47146138","saltwounds","Streaming transcription is crazy fast on an M1. Would be great to use this as a local option versus Wispr Flow.",{"id":73,"by":74,"text":75},"47149977","dSebastien","I've been using Moonshine since V1 and the results are really great. I'd say on par with Parakeet V3 while working really well with CPU only.",{"id":77,"by":78,"text":79},"47144986","armcat","This is awesome, well done guys, I’m gonna try it as my ASR component on the local voice assistant I’ve been building https:\u002F\u002Fgithub.com\u002Facatovic\u002Fova. The tiny streaming…",{"id":81,"by":82,"text":83},"47145661","aplomb1026","The streaming latency numbers are what stand out to me here. Accuracy benchmarks get all the attention, but for real-time applications (voice assistants, live captioning, in-call…",{"value":85,"source":66,"evidence":86},"Competitive with Parakeet V3 for accuracy on CPU.",[87,88],{"id":73,"by":74,"text":75},{"id":89,"by":90,"text":91},"47189303","Leftium","I think there is a streaming version of Parakeet. It is often referred to as Nemotron, though. I tried comparing Parakeet streaming with Moonshine streaming. Moonshine is…",[93,108,115],{"value":94,"source":66,"evidence":95},"Accuracy is weaker than Parakeet V3 in some tests and on streaming English.",[96,100,104],{"id":97,"by":98,"text":99},"47145321","Karrot_Kream","According to the OpenASR Leaderboard [1], looks like Parakeet V2\u002FV3 and Canary-Qwen (a Qwen finetune) handily beat Moonshine. All 3 models are open, but Parakeet is the smallest…",{"id":101,"by":102,"text":103},"47155982","SyneRyder","That was my experience when I tried Moonshine against Parakeet v3 via Handy. Moonshine was noticeably slower on my 2018-era Intel i7 PC, and didn't seem as accurate either. I'm…",{"id":105,"by":106,"text":107},"47146661","starkparker","Implemented this to transcribe voice chat in a project and the streaming accuracy in English on this was unusable, even with the medium streaming model.",{"value":109,"source":66,"evidence":110},"Non-English models are under a non-commercial license and only English weights are open.",[111],{"id":112,"by":113,"text":114},"47147794","RobotToaster","> Models for other languages are released under the Moonshine Community License, which is a non-commercial license. Weird to only release English as open weights.",{"value":116,"source":66,"evidence":117},"Claimed accuracy lacks language breakdown and clear benchmark details.",[118,122],{"id":119,"by":120,"text":121},"47145758","fareesh","Accuracy is often presumed to be english, which is fine, but it's a vague thing to say \"higher\" because does it mean higher in English only? Higher in some subset of languages?…",{"id":123,"by":124,"text":125},"47146766","heftykoo","Claiming higher accuracy than Whisper Large v3 is a bold opening move. Does your evaluation account for Whisper's notorious hallucination loops during silences (the classic…",[127,134,153],{"value":128,"source":66,"evidence":129},"Add model size to the leaderboard comparison.",[130],{"id":131,"by":132,"text":133},"47146033","reitzensteinm","Parakeet V3 is over twice the parameter count of Moonshine Medium (600m vs 245m), so it's not an apples to apples comparison. I'm actually a little surprised they haven't added…",{"value":135,"source":66,"evidence":136},"Support for browser, WebAssembly or ONNX.",[137,141,145,149],{"id":138,"by":139,"text":140},"47145241","sroussey","onnx models for browser possible?",{"id":142,"by":143,"text":144},"47145750","999900000999","Very cool. Anyway to run this in Web assembly, I have a project in mind",{"id":146,"by":147,"text":148},"47192612","Paddyz","The Parakeet vs Moonshine debate here is missing the point imo. For anything server-side or desktop, sure, Parakeet wins on raw WER. But Moonshine at 245M params vs 600M+ is a…",{"id":150,"by":151,"text":152},"47155141","fudged71","If it's using ONNX, can this be ported to Transformers.js?",{"value":154,"source":66,"evidence":155},"Include a LICENSE file for all code.",[156,160],{"id":157,"by":158,"text":159},"47144246","cyanydeez","No LICENSE no go",{"id":161,"by":162,"text":163},"47146406","mkl","The LICENSE file that refers to is missing. There's one in the python folder, but not for the rest of the code.",[165,173,184],{"value":166,"source":66,"evidence":167},"How does it compare to Parakeet?",[168,172],{"id":169,"by":170,"text":171},"47145066","g-mork","How does this compare to Parakeet, which runs wonderfully on CPU?",{"id":97,"by":98,"text":99},{"value":174,"source":66,"evidence":175},"Does it run on Raspberry Pi or phones?",[176,180],{"id":177,"by":178,"text":179},"47150237","regularfry","Oh this is fantastic. I'm most interested to see if this reaches down to the raspberry pi zero 2, because that's a whole new ballgame if it does.",{"id":181,"by":182,"text":183},"47147414","agentifysh","what is the system requirement? can it run on CPU only ? phones ? edit: This is actually amazing. I'm using handy and parakeet to type this message with just my voice. Wow.",{"value":185,"source":66,"evidence":186},"What are the streaming latency and partial stability metrics?",[187,191],{"id":188,"by":189,"text":190},"47149479","T0mSIlver","Congrats on the results. The streaming aspect is what I find most exciting here. I built a macOS dictation app (https:\u002F\u002Fgithub.com\u002FT0mSIlver\u002Flocalvoxtral) on top of Voxtral…",{"id":192,"by":193,"text":194},"47146848","guerython","Nice work. One metric I’d really like to see for streaming use cases is partial stability, not just final WER. For voice agents, the painful failure mode is partials getting…",[196,201,209],{"value":197,"source":66,"evidence":198},"Parakeet V3",[199,200],{"id":97,"by":98,"text":99},{"id":131,"by":132,"text":133},{"value":202,"source":66,"evidence":203},"Whisper",[204,208],{"id":205,"by":206,"text":207},"47146208","francislavoie","I've helped many Twitch streamers set up https:\u002F\u002Fgithub.com\u002Froyshil\u002Fobs-localvocal to plug transcription & translation into their streams, mainly for German audio to English…",{"id":123,"by":124,"text":125},{"value":210,"source":66,"evidence":211},"Handy",[212],{"id":213,"by":214,"text":215},"47146925","tuananh","Handy is amazing. Super quality app.",{"comments":40,"answered":40,"key":217},[],{"commenters":219,"max_depth":33,"maker_replies":40,"first_comment_mins":220,"lifespan_hours":16},61,37,[222,224,226,228,231,233,235,236,240,243,246,248,251,253,255,259,262,264,267,270,272,274,278,282,285,288,291,293,295,297,299,300,303,307,310,313,315,318,321,322,324,327,328,332,336,338,342,345,347,350,352,356,359,361,365,368,370,373,377,379,382,384,388,392,396,397,401,403,405,407,409,411,413,415,419,424,426,429,430,433,436,438,440,442,444,446],[97,40,32,223,98,40],"project\u002Fcritical\u002Fperformance",[131,40,40,225,132,40],"commenter\u002Fneutral\u002Fperformance",[227,40,40,225,182,40],"47147730",[229,40,40,225,230,40],"47148107","bytesandbits",[232,40,40,225,178,40],"47152809",[234,40,40,225,230,40],"47205386",[101,40,32,223,102,40],[237,40,40,238,239,40],"47146287","commenter\u002Fneutral\u002Fnone","theologic",[241,40,40,238,242,40],"47147457","jasonjmcghee",[244,40,40,225,245,40],"47146400","syntaxing",[247,40,40,225,90,40],"47189370",[249,40,40,238,250,40],"47146509","tomr75",[213,40,40,252,214,40],"commenter\u002Fpositive\u002Fnone",[254,40,40,252,182,40],"47147716",[256,40,40,257,258,40],"47148717","commenter\u002Fneutral\u002Fprivacy","alfiedotwtf",[260,40,40,257,261,40],"47148800","yorwba",[263,40,40,252,214,40],"47148925",[265,40,44,266,182,40],"47147288","project\u002Fpositive\u002Fperformance",[268,40,40,225,269,40],"47147456","remuskaos",[271,40,44,266,182,40],"47147478",[273,40,44,266,269,40],"47217138",[275,40,38,276,277,40],"47148789","project\u002Fneutral\u002Fperformance","namibj",[279,40,40,280,281,40],"47149779","other\u002Fneutral\u002Fnone","PranayKumarJain",[283,40,40,238,284,40],"47150014","cubefox",[286,40,40,238,287,40],"47148384","Dayshine",[289,40,44,266,290,40],"47150975","fittingopposite",[292,40,44,266,182,40],"47157940",[294,40,44,266,290,40],"47158773",[296,40,40,225,182,40],"47158885",[298,40,38,276,290,40],"47158978",[89,40,44,266,90,40],[301,40,40,238,302,40],"47149834","kardaj",[304,40,44,305,306,40],"47149996","project\u002Fpositive\u002Fidea","bn-usd-mistake",[308,40,40,280,309,40],"47150706","d4rkp4ttern",[311,40,40,238,312,40],"47150961","Imustaskforhelp",[314,40,40,238,309,40],"47151457",[316,40,40,238,317,40],"47155199","ValentineC",[319,40,44,320,312,40],"47151239","project\u002Fpositive\u002Fnone",[188,32,40,225,189,40],[323,32,44,266,90,40],"47188969",[205,325,38,326,206,40],2,"project\u002Fmixed\u002Fnone",[123,38,38,276,124,40],[329,18,38,330,331,40],"47145277","project\u002Fneutral\u002Fnone","asqueella",[333,44,32,334,335,40],"47145020","project\u002Fcritical\u002Fdocs","ac29",[119,337,38,276,120,40],6,[339,340,38,330,341,40],"47145735",7,"nmstoker",[343,340,38,344,90,40],"47189472","project\u002Fneutral\u002Fdocs",[112,33,32,346,113,40],"project\u002Fcritical\u002Fnone",[348,33,40,280,349,40],"47148425","riedel",[351,33,40,280,113,40],"47148712",[353,354,44,305,355,40],"47147828",9,"dagss",[357,354,40,238,358,40],"47149365","steinvakt2",[360,354,40,238,355,40],"47149888",[362,363,44,266,364,40],"47148729",10,"sourcetms",[366,363,40,238,367,40],"47149228","admiralrohan",[77,369,44,266,78,40],11,[371,372,38,330,90,40],"47177474",12,[374,375,44,266,376,40],"47148490",13,"binome",[142,378,38,330,143,40],14,[380,37,38,276,381,40],"47144580","lostmsu",[169,383,38,276,170,40],16,[385,386,44,305,387,40],"47155905",17,"Ross00781",[389,390,38,276,391,40],"47145180",18,"pzo",[393,394,38,395,387,40],"47148154",19,"project\u002Fneutral\u002Fdesign",[69,30,44,266,70,40],[398,399,38,330,400,40],"47146933",21,"oezi",[402,31,38,330,290,40],"47151025",[177,404,38,276,178,40],23,[73,406,44,266,74,40],24,[138,408,38,330,139,40],25,[105,410,32,223,106,40],26,[150,412,38,330,151,40],27,[181,414,44,266,182,40],28,[416,417,32,346,418,40],"47145739",29,"alexnewman",[420,421,38,422,423,40],"47147132",30,"maker\u002Fneutral\u002Fdocs","raybb",[157,425,32,334,158,40],31,[427,425,38,344,428,40],"47144507","bangaladore",[161,425,32,334,162,40],[431,425,40,432,277,40],"47148544","commenter\u002Fneutral\u002Fdocs",[434,425,38,344,435,40],"47144562","altruios",[192,437,38,276,193,40],32,[439,437,40,280,281,40],"47149780",[441,437,40,280,178,40],"47151415",[146,443,44,305,147,40],33,[81,445,44,266,82,40],34,[447,445,40,238,448,40],"47145826","rob",[],[451,456,461,465,469,474],{"id":452,"name":453,"points":454,"category":455},"47075124","Micasa",657,"productivity",{"id":457,"name":458,"points":459,"category":460},"47114530","CIA World Factbook Archive",496,"learning_science",{"id":462,"name":463,"points":464,"category":20},"47104667","ntransformer",395,{"id":466,"name":467,"points":468,"category":20},"47131225","Steerling-8B",328,{"id":470,"name":471,"points":472,"category":473},"47123631","PgDog",326,"infrastructure",{"id":475,"name":476,"points":477,"category":478},"47132102","X86CSS",275,"languages",false,"v2",{"id":482,"name":5,"line":6,"ai":479,"cat":20,"year":13,"points":14,"comments":15,"rec":27,"s":28,"top":18,"kw":483,"maker":11,"domain":10,"tier":19,"attempts":40,"strip":484,"url":9,"mix":485,"praise":65,"objection":94,"sub":21},47143755,[21,24,25,26],"1001000055305555055313150553535353111305",[30,31,32,33],1794036694296]