{"id":424,"date":"2026-11-12T17:13:10","date_gmt":"2026-11-12T16:13:10","guid":{"rendered":"https:\/\/machinelearning.humanativaspa.it\/en\/?p=424"},"modified":"2026-11-12T17:13:10","modified_gmt":"2026-11-12T16:13:10","slug":"tunnels-that-listen-how-ai-recognises-the-sounds-of-danger","status":"publish","type":"post","link":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/","title":{"rendered":"Tunnels that listen: how AI recognises the sounds of danger"},"content":{"rendered":"\n

Sudden braking, a crash, a cry for help. In a road tunnel the first seconds after an accident are decisive, and the first sign that something is wrong is often not something you see: it is something you hear<\/strong>. That insight is the starting point of the Sound Event Detection<\/strong> project developed by Humanativa’s Data Science team for a leading national road network operator: an artificial intelligence solution that listens to audio recorded inside tunnels, recognises danger sounds and automatically sends an alert to the control room. Here is how we got there, and why the journey is as interesting as the result.<\/p>\n\n\n\n

The challenge: spotting an alarm in a wall of noise<\/strong><\/h3>\n\n\n\n

A tunnel is one of the hardest acoustic environments imaginable. Ventilation fans, constant traffic and a reverberation lasting several seconds create a steady background noise, mostly below 2 kHz. Against this wall of sound the system has to tell apart eight types of event: braking, crashes, screams, voices, horns, sirens, tyre blow-outs and slamming doors. And it has to do so under a precise requirement: few false alarms<\/strong>, because every unnecessary alert costs operators time and attention.<\/p>\n\n\n\n

There was an even bigger hurdle: there are no real recordings of tunnel accidents to train a model on. We had to teach a machine to recognise sounds it had never “heard” in that place.<\/p>\n\n\n\n

Teaching a machine a sound it has never heard<\/strong><\/h3>\n\n\n\n

Our answer was a data preparation<\/strong> engine that builds realistic synthetic data from two ingredients: 297 event sounds from public archives and a real recording of a tunnel’s background noise. The process works like an automated audio post-production studio:<\/p>\n\n\n\n

    \n
  • Setting the scene:<\/strong> each sound receives the echo typical of a tunnel, simulated mathematically, plus a slight pitch variation, so that no two examples are identical.<\/li>\n\n\n\n
  • Mixing:<\/strong> the event is laid over the real tunnel noise at different levels (quieter than, similar to or much louder than the background) and at different moments, so the model does not learn to recognise a sound by its position.<\/li>\n\n\n\n
  • Selection:<\/strong> for each sound, 27 combinations of level, echo and pitch are tried; an automatic analysis measures how clearly the event stands out from the background, and 10 examples are kept.<\/li>\n<\/ul>\n\n\n\n

    The result is a dataset of 2,970 three-second clips<\/strong> across 8 classes (crash, braking, shouting, car door slamming, tyre burst, horn, siren, voices), fully traceable and, above all, repeatable: once real recordings become available, the dataset can grow through the very same process.<\/p>\n\n\n\n

    Looking at sound before listening to it<\/strong><\/h3>\n\n\n\n

    Before training any model, we “looked” at the data. Exploratory data analysis (EDA) turns audio into images, spectrograms, and measures energy, dynamics and frequency distribution. The findings were valuable: events are best recognised in the 2 to 8 kHz<\/strong> band, just above the area occupied by tunnel noise; 233 out of 297<\/strong> sounds clearly stand out from the background; loudness alone is not enough to separate the classes. We needed a model able to capture more complex patterns.<\/p>\n\n\n\n

    A t-SNE map of the embeddings, the internal representations produced by a pre-trained neural network, showed sound groups that were already well separated, even between acoustically similar classes such as screams and voices, or horns and sirens. It was the signal that transfer learning was the right approach.<\/p>\n\n\n\n

    Three models compared, one reasoned choice<\/strong><\/h3>\n\n\n\n

    Rather than starting from scratch, we took audio recognition models already trained on millions of sounds and specialised them on tunnel sounds: the pre-trained network extracts the signal’s features, and a small classifier on top learns to distinguish our classes. A grid search<\/strong> tested ten configurations built on four very different architectures:<\/p>\n\n\n\n

      \n
    • EfficientAT mn04 and mn10,<\/strong> lightweight convolutional networks distilled from transformer models;<\/li>\n\n\n\n
    • AST (Audio Spectrogram Transformer),<\/strong> a pure transformer with about 86 million parameters;<\/li>\n\n\n\n
    • CLAP,<\/strong> a multimodal audio-text model, the most complex of the group.<\/li>\n<\/ul>\n\n\n\n

      The selection criterion was not accuracy alone, which was close to its ceiling for almost every model, but the balance between accuracy, false alarms and simplicity<\/strong>. The winner was EfficientAT mn10 with a compact 32-neuron classifier: about 5 million parameters, able to run on a CPU, and a false alarm rate half that of the best alternatives. A concrete example of how, in applied Machine Learning, the biggest model is not always the best one.<\/p>\n\n\n\n

      From the lab to the control room<\/strong><\/h3>\n\n\n\n

      A model is only useful when it works inside a process. That is why the team engineered the entire inference chain:<\/p>\n\n\n\n

        \n
      • Flexible ingestion:<\/strong> audio files arrive from a shared folder, a REST service or a message queue.<\/li>\n\n\n\n
      • Window-based analysis:<\/strong> audio is examined in partly overlapping 5-second segments; when recognition is uncertain, the sound is classified as “other”.<\/li>\n\n\n\n
      • Alarm rules:<\/strong> an alert is raised only when a sound exceeds a configurable confidence threshold and repeats; it carries sensor, date and time, event type, confidence level and the audio excerpt to replay.<\/li>\n\n\n\n
      • Management web app:<\/strong> operators choose which sounds to monitor, with which thresholds and which control room should receive alarms, and can review history and statistics.<\/li>\n<\/ul>\n\n\n\n

        The model is exported in ONNX<\/strong> format, an open standard that keeps the production environment lightweight and independent of the framework used for training.<\/p>\n\n\n\n

        The results, read honestly<\/strong><\/h3>\n\n\n\n

        In the proof of concept the selected model reached 99% accuracy<\/strong> both in cross-validation and on an independent test set never seen during training, with high precision even on the rarest classes. A decision-threshold analysis also identified the optimal operating point, around 0.40, which minimises false alarms without missing real events.<\/p>\n\n\n\n

        These figures come from synthetic data, and we say so clearly: the natural next step is a campaign of real recordings inside tunnels to validate performance in the field. This is exactly where the work shows its value: a documented, reproducible process that is ready to learn from new data.<\/p>\n\n\n\n

        An infrastructure that listens is no longer simply an infrastructure being monitored: it becomes a system capable of detecting signals, identifying events, and providing new insights to those responsible for making decisions. This experience demonstrates that bringing artificial intelligence into real-world operations requires a methodical approach, experimentation, and a deep understanding of the context. Listening tunnels are just the beginning of a journey towards increasingly intelligent infrastructures, capable of understanding what is happening within them.<\/p>\n","protected":false},"excerpt":{"rendered":"

        In a road tunnel the first seconds after an accident are decisive, and the first sign that something is wrong is often not something you see: it is something you hear<\/p>\n","protected":false},"author":7,"featured_media":425,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[56,2,43],"tags":[124,126,127,97,125,132,57,129,130,123,128,122,131],"class_list":["post-424","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-articoli","category-approfondimenti","category-slideshow","tag-audio-ai","tag-data-augmentation","tag-deep-learning","tag-eda","tag-efficientat","tag-infrastructure","tag-mlops","tag-onnx","tag-smart-road","tag-sound-event-detection","tag-synthetic-data","tag-transfer-learning","tag-tunnel-safety"],"yoast_head":"\nTunnels that listen: how AI recognises the sounds of danger - HN Machine Learning en<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:title\" content=\"Tunnels that listen: how AI recognises the sounds of danger - HN Machine Learning en\" \/>\n<meta name=\"twitter:description\" content=\"In a road tunnel the first seconds after an accident are decisive, and the first sign that something is wrong is often not something you see: it is something you hear\" \/>\n<meta name=\"twitter:image\" content=\"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2026\/11\/HN-ML_Banner_Sound-Detection_1200x673.jpg\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"pierfrancesco\" \/>\n\t<meta name=\"twitter:label2\" content=\"Estimated reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"5 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/\"},\"author\":{\"name\":\"pierfrancesco\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#\\\/schema\\\/person\\\/d2b6fd914c90d166fceb88fea15ee8f6\"},\"headline\":\"Tunnels that listen: how AI recognises the sounds of danger\",\"datePublished\":\"2026-11-12T16:13:10+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/\"},\"wordCount\":1038,\"publisher\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/4\\\/2026\\\/11\\\/HN-ML_Banner_Sound-Detection_1200x673.jpg\",\"keywords\":[\"Audio AI\",\"Data Augmentation\",\"Deep Learning\",\"EDA\",\"EfficientAT\",\"Infrastructure\",\"MLOps\",\"ONNX\",\"Smart Road\",\"Sound Event Detection\",\"Synthetic Data\",\"Transfer Learning\",\"Tunnel Safety\"],\"articleSection\":[\"Articles\",\"Insights\",\"Slideshow\"],\"inLanguage\":\"en-GB\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/\",\"url\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/\",\"name\":\"Tunnels that listen: how AI recognises the sounds of danger - HN Machine Learning en\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/4\\\/2026\\\/11\\\/HN-ML_Banner_Sound-Detection_1200x673.jpg\",\"datePublished\":\"2026-11-12T16:13:10+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#breadcrumb\"},\"inLanguage\":\"en-GB\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#primaryimage\",\"url\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/4\\\/2026\\\/11\\\/HN-ML_Banner_Sound-Detection_1200x673.jpg\",\"contentUrl\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/4\\\/2026\\\/11\\\/HN-ML_Banner_Sound-Detection_1200x673.jpg\",\"width\":1200,\"height\":673},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Tunnels that listen: how AI recognises the sounds of danger\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#website\",\"url\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/\",\"name\":\"HN Machine Learning\",\"description\":\"\",\"publisher\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#organization\"},\"alternateName\":\"Humanativa\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-GB\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#organization\",\"name\":\"HN Machine Learning\",\"url\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/4\\\/2023\\\/09\\\/libellula_hn.jpg\",\"contentUrl\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/4\\\/2023\\\/09\\\/libellula_hn.jpg\",\"width\":696,\"height\":696,\"caption\":\"HN Machine Learning\"},\"image\":{\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/HumanativaGroupSpA\\\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/#\\\/schema\\\/person\\\/d2b6fd914c90d166fceb88fea15ee8f6\",\"name\":\"pierfrancesco\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-GB\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a5f644705853b418c243b7844d68179bf0fc5e09b724a62ce889a71ad449dc07?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a5f644705853b418c243b7844d68179bf0fc5e09b724a62ce889a71ad449dc07?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/a5f644705853b418c243b7844d68179bf0fc5e09b724a62ce889a71ad449dc07?s=96&d=mm&r=g\",\"caption\":\"pierfrancesco\"},\"url\":\"https:\\\/\\\/machinelearning.humanativaspa.it\\\/en\\\/author\\\/pierfrancesco\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Tunnels that listen: how AI recognises the sounds of danger - HN Machine Learning en","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/","twitter_card":"summary_large_image","twitter_title":"Tunnels that listen: how AI recognises the sounds of danger - HN Machine Learning en","twitter_description":"In a road tunnel the first seconds after an accident are decisive, and the first sign that something is wrong is often not something you see: it is something you hear","twitter_image":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2026\/11\/HN-ML_Banner_Sound-Detection_1200x673.jpg","twitter_misc":{"Written by":"pierfrancesco","Estimated reading time":"5 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#article","isPartOf":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/"},"author":{"name":"pierfrancesco","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#\/schema\/person\/d2b6fd914c90d166fceb88fea15ee8f6"},"headline":"Tunnels that listen: how AI recognises the sounds of danger","datePublished":"2026-11-12T16:13:10+00:00","mainEntityOfPage":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/"},"wordCount":1038,"publisher":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#organization"},"image":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#primaryimage"},"thumbnailUrl":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2026\/11\/HN-ML_Banner_Sound-Detection_1200x673.jpg","keywords":["Audio AI","Data Augmentation","Deep Learning","EDA","EfficientAT","Infrastructure","MLOps","ONNX","Smart Road","Sound Event Detection","Synthetic Data","Transfer Learning","Tunnel Safety"],"articleSection":["Articles","Insights","Slideshow"],"inLanguage":"en-GB"},{"@type":"WebPage","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/","url":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/","name":"Tunnels that listen: how AI recognises the sounds of danger - HN Machine Learning en","isPartOf":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#website"},"primaryImageOfPage":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#primaryimage"},"image":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#primaryimage"},"thumbnailUrl":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2026\/11\/HN-ML_Banner_Sound-Detection_1200x673.jpg","datePublished":"2026-11-12T16:13:10+00:00","breadcrumb":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#breadcrumb"},"inLanguage":"en-GB","potentialAction":[{"@type":"ReadAction","target":["https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/"]}]},{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#primaryimage","url":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2026\/11\/HN-ML_Banner_Sound-Detection_1200x673.jpg","contentUrl":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2026\/11\/HN-ML_Banner_Sound-Detection_1200x673.jpg","width":1200,"height":673},{"@type":"BreadcrumbList","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/tunnels-that-listen-how-ai-recognises-the-sounds-of-danger\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/machinelearning.humanativaspa.it\/en\/"},{"@type":"ListItem","position":2,"name":"Tunnels that listen: how AI recognises the sounds of danger"}]},{"@type":"WebSite","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#website","url":"https:\/\/machinelearning.humanativaspa.it\/en\/","name":"HN Machine Learning","description":"","publisher":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#organization"},"alternateName":"Humanativa","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/machinelearning.humanativaspa.it\/en\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-GB"},{"@type":"Organization","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#organization","name":"HN Machine Learning","url":"https:\/\/machinelearning.humanativaspa.it\/en\/","logo":{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#\/schema\/logo\/image\/","url":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2023\/09\/libellula_hn.jpg","contentUrl":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-content\/uploads\/sites\/4\/2023\/09\/libellula_hn.jpg","width":696,"height":696,"caption":"HN Machine Learning"},"image":{"@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/HumanativaGroupSpA\/"]},{"@type":"Person","@id":"https:\/\/machinelearning.humanativaspa.it\/en\/#\/schema\/person\/d2b6fd914c90d166fceb88fea15ee8f6","name":"pierfrancesco","image":{"@type":"ImageObject","inLanguage":"en-GB","@id":"https:\/\/secure.gravatar.com\/avatar\/a5f644705853b418c243b7844d68179bf0fc5e09b724a62ce889a71ad449dc07?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/a5f644705853b418c243b7844d68179bf0fc5e09b724a62ce889a71ad449dc07?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/a5f644705853b418c243b7844d68179bf0fc5e09b724a62ce889a71ad449dc07?s=96&d=mm&r=g","caption":"pierfrancesco"},"url":"https:\/\/machinelearning.humanativaspa.it\/en\/author\/pierfrancesco\/"}]}},"_links":{"self":[{"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/posts\/424","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/comments?post=424"}],"version-history":[{"count":1,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/posts\/424\/revisions"}],"predecessor-version":[{"id":426,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/posts\/424\/revisions\/426"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/media\/425"}],"wp:attachment":[{"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/media?parent=424"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/categories?post=424"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/machinelearning.humanativaspa.it\/en\/wp-json\/wp\/v2\/tags?post=424"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}