{"paper":{"title":"AudioPaLM: A Large Language Model That Can Speak and Listen","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Fusing a text language model with a speech model and initializing from text weights produces a system that processes and generates both modalities while outperforming prior speech translation systems.","cross_cats":["cs.AI","cs.SD","eess.AS","stat.ML"],"primary_cat":"cs.CL","authors_text":"Alexandru Tudor, Ankur Bapna, Christian Frank, Chulayuth Asawaroengchai, Dalia El Badawy, Damien Vincent, Danny Rozenberg, Dirk Padfield, Duc Dung Nguyen, Eugene Kharitonov, F\\'elix de Chaumont Quitry, Hannah Muckenhirn, James Qin, Jiahui Yu, Johan Schalkwyk, Lukas Zilka, Marco Tagliasacchi, Matt Sharifi, Michelle Tadmor Ramanovich, Mihajlo Velimirovi\\'c, Neil Zeghidour, Paul K. Rubenstein, Peter Chen, Tara Sainath, Vicky Zayats, Wei Han, Yongqiang Wang, Yu Zhang, Zal\\'an Borsos, Zhishuai Zhang","submitted_at":"2023-06-22T14:37:54Z","abstract_excerpt":"We introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models, PaLM-2 [Anil et al., 2023] and AudioLM [Borsos et al., 2022], into a unified multimodal architecture that can process and generate text and speech with applications including speech recognition and speech-to-speech translation. AudioPaLM inherits the capability to preserve paralinguistic information such as speaker identity and intonation from AudioLM and the linguistic knowledge present only in text large language models such as PaLM-2. We demons"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"The resulting model significantly outperforms existing systems for speech translation tasks and has the ability to perform zero-shot speech-to-text translation for many languages for which input/target language combinations were not seen in training.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That initializing the multimodal model with text-only LLM weights successfully transfers linguistic knowledge to speech tasks without degrading paralinguistic capabilities inherited from the speech model.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AudioPaLM unifies PaLM-2 and AudioLM to outperform prior systems on speech translation while enabling zero-shot speech-to-text for many unseen language pairs and voice transfer from short prompts.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Fusing a text language model with a speech model and initializing from text weights produces a system that processes and generates both modalities while outperforming prior speech translation systems.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"7888c1339eebf57b00fbf1ce7aa3978acdc5ffc783e49808622b74a712314bf0"},"source":{"id":"2306.12925","kind":"arxiv","version":1},"verdict":{"id":"7670d4cf-f54b-4868-bece-9f81dcfc05bb","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T07:03:23.799538Z","strongest_claim":"The resulting model significantly outperforms existing systems for speech translation tasks and has the ability to perform zero-shot speech-to-text translation for many languages for which input/target language combinations were not seen in training.","one_line_summary":"AudioPaLM unifies PaLM-2 and AudioLM to outperform prior systems on speech translation while enabling zero-shot speech-to-text for many unseen language pairs and voice transfer from short prompts.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That initializing the multimodal model with text-only LLM weights successfully transfers linguistic knowledge to speech tasks without degrading paralinguistic capabilities inherited from the speech model.","pith_extraction_headline":"Fusing a text language model with a speech model and initializing from text weights produces a system that processes and generates both modalities while outperforming prior speech translation systems."},"references":{"count":41,"sample":[{"doi":"","year":null,"title":"MusicLM: Generating Music From Text","work_id":"15e6566e-1c36-468f-966e-823248cbf87f","ref_index":1,"cited_arxiv_id":"2301.11325","is_internal_anchor":true},{"doi":"","year":null,"title":"PaLM 2 Technical Report","work_id":"905ee9a7-ea61-4a94-bd62-2600cbe3e315","ref_index":2,"cited_arxiv_id":"2305.10403","is_internal_anchor":true},{"doi":"","year":2020,"title":"ISBN 979-10-95546-34-4","work_id":"21667f47-f003-4382-b396-17b57c530dd7","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"mSLAM: Massively multilingual joint pre-training for speech and text","work_id":"c143c52c-596d-4470-b0f4-abac681bd925","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2019,"title":"L. Barrault, O. Bojar, M. R. Costa-jussà, C. Federmann, M. Fishel, Y . Graham, B. Haddow, M. Huck, P. Koehn, S. Malmasi, C. Monz, M. Müller, S. Pal, M. Post, and M. Zampieri. Findings of the 2019 conf","work_id":"0c91c9d2-37ba-46d7-b2b3-d74c9b276fd3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":41,"snapshot_sha256":"0ff9b9a65c23f886b56b16c49d9f7962648b154f4248d33ff4705ac2813f17be","internal_anchors":10},"formal_canon":{"evidence_count":2,"snapshot_sha256":"527b2d1c735814595e76ae9d9cc367913104b5e6ce7bee41f254d58906f5cafa"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}