[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83630-en":3,"doc-seo-83630-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},83630,1649267921044,"Ava Thompson","https://us-avatar.wpscdn.com/avatar/1800007509477c92dfb?_k=1782875107921204101",8,"Research & Report","Speaker Head Orientation Estimation with a Single Microphone Array Using Phase Spectrogram Features","Estimating a speaker’s head orientation from audio enables richer human–machine interaction in smart environments, meetings, and driver monitoring. The study introduces a deep neural network that uses the phase component of the short-time Fourier transform (STFT) from a single microphone array. The architecture integrates convolutional, recurrent, and self-attention layers. Pre-training on a large simulated dataset built from voice directivity patterns and fine-tuning on real recordings yields state-of-the-art results in both clean and noisy conditions, while personalization reaches a mean angular error of 11.3°.","SPEAKER HEAD ORIENTATION ESTIMATION WITH A SINGLE MICROPHONE ARRAY  \nUSING PHASE SPECTROGRAM FEATURES  \nB ´alint Turi, Archontis Politis, Parthasaarathy Sudarsanam, Tuomas Virtanen  \nAudio Research Group, Tampere University, Finland  \narXiv :2607 .02 129v 1 [ cs . SD] 2 Jul 2026  \nABSTRACT  \nEstimating a speaker’s head orientation from audio can provide valuable information in smart environments, meetings, and driver monitoring. We propose a novel approach that leverages the phase component of the short-time Fourier transform from a single microphone array as input to a deep neural network combining convolutional, recurrent, and selfattention layers. Unlike prior methods that use physicsinformed handcrafted features or raw waveform inputs, our approach enables robust learning from simulated and real data. Trained on a large-scale dataset generated with voice directivity patterns and fine-tuned on real recordings, our model achieves state-of-the-art accuracy, outperforming baselines under both clean and noisy conditions. Personalization experiments further demonstrate significant gains, reaching a mean angular error of 11.3° when adapting to individual users and environments.  \nIndex Terms— Head orientation, phase spectrogram, speech processing  \n1. INTRODUCTION  \nEstimating a speaker’s head orientation has become increasingly important in modern human–machine interaction, as it provides cues about attention, intent, and conversational context. For example, in smart home environments, orientation information can disambiguate which device a user’s command is directed to [1, 2, 3] . In meeting rooms, it can help determine who a speaker is talking to, enabling addressee recognition and improved transcription [4, 5] . In vehicles, it can support driver monitoring systems by tracking attention, contributing to road safety [6] .  \nWhile vision-based estimation of head orientation has been extensively studied [7], they require calibrated cameras, are sensitive to occlusions and lighting, and raise privacy concerns. These challenges motivate audio-based solutions that are accurate and more privacy-preserving, leveraging the microphones already embedded in smart speakers, conferencing platforms, and vehicles. Human speech production is inherently directional, as the vocal tract radiates sound unevenly, producing systematic variations in spectral coloration, inter-channel phase differences, and reflection patterns. By exploiting this directional property of speech, audio-only sys-  \ntems can provide a practical and effective means for head orientation estimation.  \nEarly research on acoustic head orientation estimation relied on model-based approaches and large microphone arrays. For example, [8] employed a Huge Microphone Array (HMA) with 448 microphones, while similar HMA-based systems [9, 10] estimated orientation from low-pass filtered signal energy differences. Other studies [11, 12] used GCCPHAT [13] to extract cross-correlation peaks between microphone pairs, and additional work [14, 15] investigated highto-low frequency band energy ratios [14] . Although these approaches required minimal training, they were impractical for real-world deployment due to large hardware requirements, computational cost, and sensitivity to noise. Later efforts reduced hardware complexity, with [16] using six arrays and [2] using two four-channel arrays; however, they still assumed known array geometry and user position. In contrast, single compact devices with small embedded arrays are more aligned with practical applications.  \nMore recently, machine learning approaches have been explored. [17] proposed a hidden Markov model using singlechannel audio, though with limited performance compared to multichannel methods. Extensions based on phase coefficients of the cross-power spectrum [18] were adapted from loudspeaker localization and remained noise sensitive. The work in [3] predicted eight discrete orientations using a decision tree trained on spectr","cbCaipNNGGoVAYKc","https://ap.wps.com/l/cbCaipNNGGoVAYKc","pdf",677506,1,5,"English","en",105,"# Abstract\n# Introduction\n# Problem Definition and Methodology","[{\"question\":\"What input feature does the proposed method use for head orientation estimation?\",\"answer\":\"It uses the phase component of the short-time Fourier transform (STFT) extracted from a single microphone array.\"},{\"question\":\"How is the neural network structured?\",\"answer\":\"The model combines convolutional layers with recurrent layers and self-attention layers to learn orientation-relevant representations.\"},{\"question\":\"What training strategy improves performance on real recordings?\",\"answer\":\"The approach pre-trains on a large simulated dataset generated using voice directivity patterns, then fine-tunes the model on real audio recordings.\"}]",1784189386,13,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"speaker-head-orientation-estimation-with-a-single-microphone-array-using-phase-spectrogram-features","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/speaker-head-orientation-estimation-with-a-single-microphone-array-using-phase-spectrogram-features/83630/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What input feature does the proposed method use for head orientation estimation?","Question",{"text":75,"@type":76},"It uses the phase component of the short-time Fourier transform (STFT) extracted from a single microphone array.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How is the neural network structured?",{"text":80,"@type":76},"The model combines convolutional layers with recurrent layers and self-attention layers to learn orientation-relevant representations.",{"name":82,"@type":73,"acceptedAnswer":83},"What training strategy improves performance on real recordings?",{"text":84,"@type":76},"The approach pre-trains on a large simulated dataset generated using voice directivity patterns, then fine-tunes the model on real audio recordings.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,109,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":21,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":21,"slug":137},19,"General","general"]