[{"data":1,"prerenderedAt":298},["ShallowReactive",2],{"project-49699267":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":29,"sentiment":30,"stance_mix":31,"scored_comments":36,"aspects":37,"strip":46,"post_text":48,"what":49,"said":81,"maker_in_thread":167,"shape":179,"grid":183,"attempts":259,"same_week":260,"maker_shows":41,"maker_hidden":290,"reader_version":291,"card":292},"49699267","Nari Qwen3-TTS and Qwen3-ASR","High accuracy, low latency and cost","title","Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost","https:\u002F\u002Fnarilabs.com\u002Fblog\u002Fnari-labs-leads-coval-voice-ai-benchmarks\u002F","narilabs.com","toebee",1789402078,2026,92,31,742,4366,17,"notable","ai","speech models","api",[24,25,26,27,28],"text to speech","speech recognition","qwen3-tts","low latency inference","open source tts","debated",-0.097,{"positive":32,"neutral":33,"mixed":34,"critical":35},5,4,0,7,16,{"idea":38,"performance":40,"execution":42,"pricing":43,"docs":44,"design":45},{"positive":39,"critical":34},1,{"positive":41,"critical":39},2,{"positive":34,"critical":41},{"positive":34,"critical":41},{"positive":34,"critical":39},{"positive":34,"critical":39},[47,47,34,32,34,34,34,39,34,32,34,39,34,39,39,34,47,32,47,34,32,34,34,39,34,34,32,34,39,34,39],3,"Hey HN, Toby from Nari Labs here.\n\nWe've been working on making OSS speech models super-fast. Last year, we built Dia, the first OSS text-to-speech model capable of doing natural dialogue. Since then, so many more great speech models have been released to the public.\n\nBut the market is still dominated by closed source models. We think that's an inference problem. Existing systems such as vLLM \u002F SGLang are not well suited for multimodal inference. To prove this, we built an inference engine specialized for Qwen3-TTS and open-sourced it (https:\u002F\u002Fgithub.com\u002Fnari-labs\u002Fnari-qwen3-tts). Running at s",{"problem":50,"audience":56,"pricing":57,"motivation":64,"time_to_build":56,"stack":69,"openness":78,"stage":79,"team_size":80},{"value":51,"source":52,"evidence":53},"Closed-source speech models dominate; open models need faster, cheaper serving","maker",[54],{"id":55},"post",null,{"value":58,"source":52,"evidence":59},"Cheapest endpoint on Coval benchmark, pricing aims to drop further",[60,61],{"id":55},{"id":62,"by":11,"text":63},"49710422","if we can lower the pricing by not 50% but 10x, then I think it would be something people want. we are taking the bet that OSS models will take a huge chunk of market share not…",{"value":65,"source":52,"evidence":66},"Make open speech models fast and cheap so closed models lose share",[67,68],{"id":55},{"id":62,"by":11,"text":63},[70,74],{"value":71,"source":52,"evidence":72},"Qwen3-TTS",[73],{"id":55},{"value":75,"source":52,"evidence":76},"Qwen3-ASR",[77],{"id":55},"open_source","launched","unknown",{"praise":82,"objections":95,"requests":125,"questions":144,"alternatives":149},[83],{"value":84,"source":85,"evidence":86},"Impressive open-source speech performance and cost","commenter",[87,91],{"id":88,"by":89,"text":90},"49707098","karimf","This is awesome. Thanks for pushing the audio pareto frontier forward. Probably far fetched for now, but I think the next big evolution is building the pareto\u002Fmuch cheaper…",{"id":92,"by":93,"text":94},"49706674","apimade","https:\u002F\u002Fapimade.com\u002Faudio-compare.html Added it to my blind TTS model comparison leaderboard. So far Darwin TTS is the open model leading the pack, ElevenLabs is at the lead.",[96,107,114],{"value":97,"source":85,"evidence":98},"Voice switches or speed issues in generated audio",[99,103],{"id":100,"by":101,"text":102},"49701701","rahimnathwani","For some reason it switched voices half way through a 33 second clip. For OP the clip name is nari-nina-01a0a12f-980a-765e-8029-fa56bd23210d.wav",{"id":104,"by":105,"text":106},"49704759","konart","All TTS generations are too fast. It's almost I'm listening to a podcast on 1.25-1.5x speed.",{"value":108,"source":85,"evidence":109},"Claims need independent evaluation",[110],{"id":111,"by":112,"text":113},"49702361","yoloakki","You definitely need independent evals by Datapoint AI or someone who can verify your claims about TTS quality",{"value":115,"source":85,"evidence":116},"Market is crowded and changing fast",[117,121],{"id":118,"by":119,"text":120},"49714358","recentlypostedj","Question: How do you plan to differentiate, because there are so many TTS and its constantly changing every month who would become better",{"id":122,"by":123,"text":124},"49702762","iharnoor","By next month the competition for TTS will be even more! Voice models are not winner take all market unlike LLM APIs Coming here as Developer Relations at AssemblyAI",[126,134],{"value":127,"source":85,"evidence":128},"Demos and independent benchmarks",[129,133],{"id":130,"by":131,"text":132},"49701631","ipsum2","If you're going to announce a TTS model, service, or whatever, you really need demos.",{"id":111,"by":112,"text":113},{"value":135,"source":85,"evidence":136},"Open-source Qwen3-ASR inference engine",[137,141],{"id":138,"by":139,"text":140},"49701336","meatmanek","> and Qwen3-ASR Is the ASR inference engine open source as well?",{"id":142,"by":139,"text":143},"49705250","I meant the Nari inference engine for Qwen3-ASR. I'm aware that Qwen3-ASR is open source, but I don't see a repo under https:\u002F\u002Fgithub.com\u002Fnari-labs for nari-qwen3-asr or similar.…",[145],{"value":146,"source":85,"evidence":147},"How will Nari differentiate in a crowded TTS market",[148],{"id":118,"by":119,"text":120},[150,154,161],{"value":151,"source":85,"evidence":152},"ElevenLabs",[153],{"id":92,"by":93,"text":94},{"value":155,"source":85,"evidence":156},"Darwin TTS",[157,158],{"id":92,"by":93,"text":94},{"id":159,"by":93,"text":160},"49707061","Darwin TTS is based on Qwen3-TTS https:\u002F\u002Fhuggingface.co\u002Fzeropointnine\u002FDarwin-TTS-1.7B-Cross-Qwen3Tokenizer",{"value":162,"source":52,"evidence":163},"NVIDIA PersonaPlex",[164],{"id":165,"by":11,"text":166},"49708055","agreed. we've been doing some work around NVIDIA personaplex 7b, but its quality is quite far from GPT-Live-1, esp in terms of intelligence. Once a good OSS model is out, we'll…",{"comments":168,"answered":41,"key":169},10,[170,173,176],{"id":171,"text":172},"49705744","thanks for the interest! we have a blog post on exactly how we did it: https:\u002F\u002Fnarilabs.com\u002Fblog\u002Fqwen3-tts-speed-cost-frontier\u002F",{"id":174,"text":175},"49707989","awesome! will look into Darwin TTS. super interesting",{"id":177,"text":178},"49705760","hey, thanks for letting us know! will look into the issue and see what went wrong.",{"commenters":180,"max_depth":32,"maker_replies":168,"first_comment_mins":181,"lifespan_hours":182},19,106,26,[184,186,190,192,194,198,199,201,203,205,207,208,212,214,216,217,219,222,225,226,228,231,234,236,239,243,245,248,250,253,255],[118,34,47,185,119,34],"project\u002Fneutral\u002Fidea",[187,39,47,188,189,34],"49701055","project\u002Fneutral\u002Fperformance","asaiacai",[171,39,34,191,11,39],"maker\u002Fneutral\u002Fdocs",[92,41,32,193,93,34],"project\u002Fpositive\u002Fperformance",[195,41,34,196,197,34],"49706837","other\u002Fneutral\u002Fnone","barney54",[159,41,34,196,93,34],[174,41,34,200,11,39],"maker\u002Fpositive\u002Fnone",[100,47,39,202,101,34],"project\u002Fcritical\u002Fexecution",[177,47,34,204,11,39],"maker\u002Fneutral\u002Fexecution",[88,33,32,206,89,34],"project\u002Fpositive\u002Fidea",[165,33,34,185,11,39],[209,33,39,210,211,34],"49708074","project\u002Fcritical\u002Fpricing","6879346626",[62,33,34,213,11,39],"maker\u002Fneutral\u002Fpricing",[215,33,39,210,211,34],"49716707",[104,32,39,202,105,34],[218,32,34,204,11,39],"49705734",[138,220,47,221,139,34],6,"project\u002Fneutral\u002Fnone",[223,220,32,193,224,34],"49702829","nshm",[142,220,47,191,139,34],[227,220,34,191,11,39],"49705783",[229,220,32,200,230,34],"49705947","kshmir",[232,220,34,196,233,34],"49703223","verdverm",[122,35,34,235,123,34],"other\u002Fneutral\u002Fidea",[111,237,39,238,112,34],8,"project\u002Fcritical\u002Fperformance",[240,241,34,196,242,34],"49702076",9,"mowmiatlas",[244,241,34,200,11,39],"49705756",[246,168,32,200,247,34],"49702667","DylanMerigaud",[249,168,34,200,11,39],"49708058",[130,251,39,252,131,34],11,"project\u002Fcritical\u002Fdocs",[254,251,34,191,11,39],"49705727",[256,251,39,257,258,34],"49708412","project\u002Fcritical\u002Fdesign","emayljames",[],[261,266,271,276,281,285],{"id":262,"name":263,"points":264,"category":265},"49711544","Fugleramme",2391,"creative",{"id":267,"name":268,"points":269,"category":270},"49637385","rivendell.dmitrybrant.com",633,"learning_science",{"id":272,"name":273,"points":274,"category":275},"49693971","Inflight Simulator",446,"games",{"id":277,"name":278,"points":279,"category":280},"49712278","Capsule",380,"dev_tools",{"id":282,"name":283,"points":284,"category":270},"49649789","Bodily Oddities",341,{"id":286,"name":287,"points":288,"category":289},"49712102","modem-thing",210,"hardware",false,"v2",{"id":293,"name":5,"line":6,"ai":290,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":294,"top":18,"kw":295,"maker":11,"domain":10,"tier":19,"attempts":34,"strip":296,"url":9,"mix":297,"praise":84,"objection":97,"sub":21},49699267,-0.1,[24,25,26,27,28],"3305000105010110353050010050101",[32,33,34,35],1794036710068]