<?xml version="1.0" encoding="utf-8"?>
<export-example>
  <doc>
    <id>4351</id>
    <completedYear>2025</completedYear>
    <publishedYear>2025</publishedYear>
    <thesisYearAccepted/>
    <language>eng</language>
    <pageFirst>1</pageFirst>
    <pageLast>5</pageLast>
    <pageNumber>5</pageNumber>
    <edition/>
    <issue/>
    <volume/>
    <type>conferenceobject</type>
    <publisherName>IEEE</publisherName>
    <publisherPlace/>
    <creatingCorporation/>
    <contributingCorporation/>
    <belongsToBibliography>1</belongsToBibliography>
    <completedDate>--</completedDate>
    <publishedDate>--</publishedDate>
    <thesisDateAccepted>--</thesisDateAccepted>
    <title language="eng">SELMA: A Speech-Enabled Language Model for Virtual Assistant Interactions</title>
    <abstract language="eng">In this work, we present and evaluate SELMA, a Speech-Enabled Language Model for virtual Assistant interactions that integrates audio and text as inputs to a Large Language Model (LLM). SELMA is designed to handle three primary and two auxiliary tasks related to interactions with virtual assistants simultaneously within a single end-to-end model. We employ low-rank adaptation modules for parameter-efficient training of both the audio encoder and the LLM. Additionally, we implement a feature pooling strategy enabling the system to recognize global patterns and improve accuracy on tasks less&#13;
reliant on individual sequence elements. Experimental results on Voice&#13;
Trigger (VT) detection, Device-Directed Speech Detection (DDSD), and&#13;
Automatic Speech Recognition (ASR), demonstrate that our approach both simplifies the typical input processing pipeline of virtual assistants significantly and also improves performance compared to dedicated models for each individual task. SELMA yields relative Equal-Error Rate improvements of 64% on the VT detection task, and 22% on DDSD, while also achieving word error rates close to the baseline.</abstract>
    <parentTitle language="eng">ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)</parentTitle>
    <identifier type="doi">10.1109/ICASSP49660.2025.10890139</identifier>
    <identifier type="issn">2379-190X</identifier>
    <enrichment key="opus_doi_flag">true</enrichment>
    <enrichment key="opus_import_data">{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,9,19]],"date-time":"2025-09-19T09:13:19Z","timestamp":1758273199068,"version":"3.41.2"},"reference-count":39,"publisher":"IEEE","license":[{"start":{"date-parts":[[2025,4,6]],"date-time":"2025-04-06T00:00:00Z","timestamp":1743897600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,4,6]],"date-time":"2025-04-06T00:00:00Z","timestamp":1743897600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,4,6]]},"DOI":"10.1109\/icassp49660.2025.10890139","type":"proceedings-article","created":{"date-parts":[[2025,3,12]],"date-time":"2025-03-12T17:15:19Z","timestamp":1741799719000},"page":"1-5","source":"Crossref","is-referenced-by-count":1,"title":["SELMA: A Speech-Enabled Language Model for Virtual Assistant Interactions"],"prefix":"10.1109","author":[{"given":"Dominik","family":"Wagner","sequence":"first","affiliation":[{"name":"TH N&amp;#x00FC;rnberg"}]},{"given":"Alexander","family":"Churchill","sequence":"additional","affiliation":[{"name":"Apple"}]},{"given":"Siddharth","family":"Sigtia","sequence":"additional","affiliation":[{"name":"Apple"}]},{"given":"Erik","family":"Marchi","sequence":"additional","affiliation":[{"name":"Apple"}]}],"member":"263","reference":[{"volume-title":"Voice trigger system for Siri","year":"2023","author":"Team","key":"ref1"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2018-2204"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSPW62465.2024.10626172"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2020-1491"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2022-529"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2015-352"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1109\/ASRU.2017.8268946"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP49357.2023.10096976"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP48485.2024.10446224"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2012-83"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2018-1531"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2022-11228"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2021-1428"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP40776.2020.9053781"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP.2019.8682617"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP.2019.8683565"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP49357.2023.10096184"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2019-2840"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1109\/icassp48485.2024.10446421"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2024-1361"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/CVPRW59228.2023.00490"},{"article-title":"Apple intelligence foundation language models","year":"2024","author":"Gunter","key":"ref22"},{"article-title":"Phi-3 technical report: A highly capable language model locally on your phone","year":"2024","author":"Abdin","key":"ref23"},{"key":"ref24","first-page":"22199","article-title":"Large language models are zero-shot reasoners","volume":"35","author":"Kojima","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"article-title":"Language models are few-shot learners","year":"2020","author":"Brown","key":"ref25"},{"article-title":"Audio flamingo: A novel audio language model with few-shot learning and dialogue abilities","year":"2024","author":"Kong","key":"ref26"},{"article-title":"SALMONN: Towards generic hearing abilities for large language models","volume-title":"The Twelfth International Conference on Learning Representations","author":"Tang","key":"ref27"},{"article-title":"Qwen-audio: Advancing universal audio understanding via unified large-scale audio-language models","year":"2023","author":"Chu","key":"ref28"},{"article-title":"Audiopalm: A large language model that can speak and listen","year":"2023","author":"Rubenstein","key":"ref29"},{"key":"ref30","first-page":"18090","article-title":"Pengi: An audio language model for audio tasks","volume":"36","author":"Deshmukh","year":"2023","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref31","first-page":"28","article-title":"Coding dialogs with the damsl annotation scheme","volume-title":"AAAI Fall Symposium on Communicative Action in Humans and Machines","volume":"56","author":"Core"},{"article-title":"Multimodal Data and Resource Efficient Device-Directed Speech Detection with Large Foundation Models","volume-title":"Third Workshop on Efficient Natural Language and Speech Processing (ENLSP-III) at NeurIPS 2023","author":"Wagner","key":"ref32"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.48550\/ARXIV.1706.03762"},{"article-title":"Robust speech recognition via large-scale weak supervision","year":"2022","author":"Radford","key":"ref34"},{"article-title":"Qwen technical report","year":"2023","author":"Bai","key":"ref35"},{"key":"ref36","first-page":"75067","article-title":"Quantizable transformers: Removing outliers by helping attention heads do nothing","volume":"36","author":"Bondarenko","year":"2023","journal-title":"Advances in Neural Information Processing Systems"},{"article-title":"LoRA: Low-rank adaptation of large language models","volume-title":"International Conference on Learning Representations","author":"Hu","key":"ref37"},{"article-title":"Decoupled weight decay regularization","volume-title":"International Conference on Learning Representations","author":"Loshchilov","key":"ref38"},{"article-title":"Phi-3 technical report: A highly capable language model locally on your phone","year":"2024","author":"Abdin","key":"ref39"}],"event":{"name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","start":{"date-parts":[[2025,4,6]]},"location":"Hyderabad, India","end":{"date-parts":[[2025,4,11]]}},"container-title":["ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10887540\/10887541\/10890139.pdf?arnumber=10890139","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,7,26]],"date-time":"2025-07-26T06:46:11Z","timestamp":1753512371000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/10890139\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,4,6]]},"references-count":39,"URL":"https:\/\/doi.org\/10.1109\/icassp49660.2025.10890139","relation":{},"subject":[],"published":{"date-parts":[[2025,4,6]]}}}</enrichment>
    <enrichment key="local_crossrefDocumentType">proceedings-article</enrichment>
    <enrichment key="local_crossrefLicence">https://doi.org/10.15223/policy-029</enrichment>
    <enrichment key="local_import_origin">crossref</enrichment>
    <enrichment key="local_doiImportPopulated">PersonAuthorFirstName_1,PersonAuthorLastName_1,PersonAuthorFirstName_2,PersonAuthorLastName_2,PersonAuthorFirstName_3,PersonAuthorLastName_3,PersonAuthorFirstName_4,PersonAuthorLastName_4,Enrichmentconference_title,Enrichmentconference_place,PublisherName,TitleMain_1,TitleParent_1,PageNumber,PageFirst,PageLast,CompletedYear,Enrichmentlocal_crossrefLicence</enrichment>
    <enrichment key="conference_title">ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)</enrichment>
    <enrichment key="conference_place">Hyderabad, India</enrichment>
    <enrichment key="Reviewstatus">Begutachtet/Reviewed</enrichment>
    <enrichment key="opus.source">doi-import</enrichment>
    <enrichment key="opus.doi.autoCreate">false</enrichment>
    <enrichment key="opus.urn.autoCreate">false</enrichment>
    <author>Dominik Wagner</author>
    <author>Alexander Churchill</author>
    <author>Siddharth Sigtia</author>
    <author>Erik Marchi</author>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>multi-task</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>multimodal</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>virtual assistant</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>large language</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>model</value>
    </subject>
    <subject>
      <language>eng</language>
      <type>uncontrolled</type>
      <value>low-rank adaptation</value>
    </subject>
    <collection role="institutes" number="">Fakultät Informatik</collection>
    <collection role="Forschungsschwerpunkt" number="5">Digitalisierung &amp; Künstliche Intelligenz</collection>
    <collection role="institutes" number="">Zentrum für Künstliche Intelligenz (KIZ)</collection>
    <file>https://opus4.kobv.de/opus4-ohm/files/4351/ICASSP25_wagner_selma_accepted.pdf</file>
  </doc>
</export-example>
