[{"data":1,"prerenderedAt":275},["ShallowReactive",2],{"project-43283317":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":35,"aspects":36,"strip":42,"post_text":44,"what":45,"said":78,"maker_in_thread":157,"shape":165,"grid":169,"attempts":237,"same_week":238,"maker_shows":43,"maker_hidden":267,"reader_version":268,"card":269},"43283317","Smart Turn","Open-source, native audio turn detection model","title","Show HN: Open-source, native audio turn detection model","https:\u002F\u002Fgithub.com\u002Fpipecat-ai\u002Fsmart-turn","github.com","kwindla",1741285248,2025,126,28,593,3837,16,"notable","ai","audio turn detection model","library",[24,25,26,27,28],"turn detection","voice ai","audio model","endpoint detection","open source","loved",0.368,{"positive":32,"neutral":32,"mixed":33,"critical":34},7,0,1,15,{"idea":37,"docs":39,"design":40},{"positive":38,"critical":33},5,{"positive":33,"critical":33},{"positive":41,"critical":34},2,[38,33,33,43,33,43,38,33,33,33,33,33,43,43,33,33,43,38,38,33,38,34,38,33,38,43,33,43],3,"Our goal with this project is to build a completely open source, state of the art turn detection model that can be used in any voice AI application.\n\nI've been experimenting with LLM voice conversations since GPT-4 was first released. (There's a previous front page Show HN about Pipecat, the open source voice AI orchestration framework I work on. [1])\n\nIt's been almost two years, and for most of that time, I've been expecting that someone would \"solve\" turn detection. We all built initial, pretty good 80\u002F20 versions of turn detection on top of VAD (voice activity detection) models. And then, a",{"problem":46,"audience":55,"pricing":59,"motivation":60,"time_to_build":59,"stack":64,"openness":75,"stage":76,"team_size":77},{"value":47,"source":48,"evidence":49},"Detecting when a speaker has finished talking so an LLM voice agent can respond","maker",[50,52],{"id":51},"post",{"id":53,"by":11,"text":54},"43287164","Turn detection is deciding when a person has finished talking and expects the other party in a conversation to respond. In this case, the other party in the conversation is an LLM!",{"value":56,"source":48,"evidence":57},"Voice AI application developers and audio ML engineers",[58],{"id":51},null,{"value":61,"source":48,"evidence":62},"Maker expected someone would solve turn detection and found the ecosystem stuck, so built an open model",[63],{"id":51},[65],{"value":66,"source":48,"evidence":67},"Wav2Vec2-BERT",[68,72],{"id":69,"by":70,"text":71},"43287351","foundzen","I got most of my answers from the README. Well written. I read most of it. Can you share what kind of resources (and how much of them) were required to fine tune Wav2Vec2-BERT?",{"id":73,"by":11,"text":74},"43287423","It takes about 45 minutes to do the current training run on an L4 GPU with these settings: ``` # Training parameters \"learning_rate\": 5e-5, \"num_epochs\": 10, \"train_batch_size\":…","open_source","beta","team",{"praise":79,"objections":99,"requests":114,"questions":129,"alternatives":148},[80,85,92],{"value":81,"source":82,"evidence":83},"Model is well documented and the README is clear","commenter",[84],{"id":69,"by":70,"text":71},{"value":86,"source":82,"evidence":87},"Architecture matches existing turn-based models",[88],{"id":89,"by":90,"text":91},"43286375","written-beyond","Having reviewed a few turn based models your implementation is pretty inline with them. Excited to see how this matures!",{"value":93,"source":82,"evidence":94},"Important problem for voice agents",[95],{"id":96,"by":97,"text":98},"43287629","xp84","I'm excited to see this particular technology developing more. From the absolute worst speech systems such as Siri, who will happily interrupt to respond with nonsense at the…",[100,107],{"value":101,"source":82,"evidence":102},"Turn detection may not be solvable without push-to-talk",[103],{"id":104,"by":105,"text":106},"43288257","pzo","I will have a look at this. Played with pipecat before and it's great, switched to sherpa-onnx though since I need something that compile to native and can run on edge devices.…",{"value":108,"source":82,"evidence":109},"Model could interrupt users inappropriately",[110],{"id":111,"by":112,"text":113},"43291082","meltyness","... but will the model learn when to interrupt you out of frustration with your ongoing statements, and start shouting? it seems like for the obvious use-cases there might need…",[115,122],{"value":116,"source":82,"evidence":117},"Support for multiple speakers",[118],{"id":119,"by":120,"text":121},"43287814","lostmsu","Does this support multiple speakers?",{"value":123,"source":82,"evidence":124},"Better SEO and explanation of the term turn detection",[125],{"id":126,"by":127,"text":128},"43296042","mncharity","Re SEO, I note \"utterance\" only occurs once, in a perhaps-ephemeral \"Things to do\" description. To help with \"what is?\" and SEO, perhaps something like \"Turn detection (aka…",[130,137,144],{"value":131,"source":82,"evidence":132},"What is turn detection?",[133],{"id":134,"by":135,"text":136},"43287130","remram","Ok what's turn detection?",{"value":138,"source":82,"evidence":139},"How many parameters does the model have?",[140],{"id":141,"by":142,"text":143},"43286115","zamalek","As an [diagnosed] HF autistic person, this is unironically something I would go for in an earpiece. How many parameters is the model?",{"value":145,"source":82,"evidence":146},"What resources were needed to fine tune the model?",[147],{"id":69,"by":70,"text":71},[149,153],{"value":150,"source":82,"evidence":151},"sherpa-onnx",[152],{"id":104,"by":105,"text":106},{"value":154,"source":48,"evidence":155},"LiveKit text-based turn detection model",[156],{"id":51},{"comments":32,"answered":158,"key":159},4,[160,163,164],{"id":161,"text":162},"43287516","A couple of interesting updates today: - 100ms inference using CoreML: https:\u002F\u002Fx.com\u002Fmaxxrubin_\u002Fstatus\u002F1897864136698347857 - An LSTM model (1\u002F7th the size) trained on a subset of…",{"id":73,"text":74},{"id":53,"text":54},{"commenters":166,"max_depth":38,"maker_replies":32,"first_comment_mins":167,"lifespan_hours":168},18,315,38,[170,172,176,178,180,182,184,187,188,191,194,196,199,203,206,208,210,211,212,213,216,220,222,224,226,229,232,234],[104,33,38,171,105,33],"project\u002Fpositive\u002Fidea",[173,33,33,174,175,33],"43298561","other\u002Fneutral\u002Fnone","genewitch",[161,34,33,177,11,34],"project\u002Fpositive\u002Fperformance",[69,41,43,179,70,33],"maker\u002Fneutral\u002Fdocs",[73,41,33,181,11,34],"maker\u002Fneutral\u002Fexecution",[134,43,43,183,135,33],"project\u002Fneutral\u002Fnone",[185,43,38,171,186,33],"43287163","ry167",[53,43,33,179,11,34],[189,43,33,190,135,33],"43287185","commenter\u002Fneutral\u002Fnone",[192,43,33,190,193,33],"43291887","whiddershins",[195,43,33,190,135,33],"43292557",[197,43,33,174,198,33],"43294913","password4321",[200,43,43,201,202,33],"43287779","project\u002Fneutral\u002Fdocs","woodson",[204,43,43,201,205,33],"43291503","lelag",[207,43,33,179,11,34],"43292289",[209,43,33,190,205,33],"43292804",[126,43,43,179,127,33],[96,158,38,171,97,33],[141,38,38,171,142,33],[214,38,33,215,11,34],"43286789","maker\u002Fneutral\u002Fdesign",[217,38,38,218,219,33],"43287524","project\u002Fpositive\u002Fdesign","cyberbiosecure",[111,38,34,221,112,33],"project\u002Fcritical\u002Fdesign",[89,223,38,218,90,33],6,[225,223,33,215,11,34],"43286791",[227,32,38,171,228,33],"43287556","prophesi",[119,230,43,231,120,33],8,"project\u002Fneutral\u002Fdesign",[233,230,33,215,11,34],"43292316",[235,236,43,183,219,33],"43287520",9,[],[239,243,248,253,258,262],{"id":240,"name":241,"points":242,"category":20},"43331582","Factorio Learning Environment",749,{"id":244,"name":245,"points":246,"category":247},"43328095","Seven39",735,"consumer_social",{"id":249,"name":250,"points":251,"category":252},"43347306","Time Portal",497,"games",{"id":254,"name":255,"points":256,"category":257},"43280615","CodeTracer",334,"dev_tools",{"id":259,"name":260,"points":261,"category":247},"43304785","daale.club",302,{"id":263,"name":264,"points":265,"category":266},"43333661","Pila",301,"hardware",false,"v2",{"id":270,"name":5,"line":6,"ai":267,"cat":20,"year":13,"points":14,"comments":15,"rec":29,"s":271,"top":18,"kw":272,"maker":11,"domain":10,"tier":19,"attempts":33,"strip":273,"url":9,"mix":274,"praise":81,"objection":101,"sub":21},43283317,0.37,[24,25,26,27,28],"5003035000003300355051505303",[32,32,33,34],1794036670797]