--- title: Language Discrimination Improves Linguistic Learning in... description: Apple researchers present an experiment showing that multilingual self-supervised speech models (based on HuBERT) trained on English/French data underperform... canonical: https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Language Discrimination Improves Linguistic Learning in Multilingual Speech Models | daily.dev og:description: Apple researchers present an experiment showing that multilingual self-supervised speech models (based on HuBERT) trained on English/French data underperform... og:url: https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump og:image: https://api.daily.dev/og/posts/sZJjQ0umP.png og:image:alt: Language Discrimination Improves Linguistic Learning in Multilingual Speech Models og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Language Discrimination Improves Linguistic Learning in Multilingual Speech Models **[Apple Machine Learning Research](https://daily.dev/sources/mlapple)** · 1 min read · 0 upvotes · 0 comments ## Summary Apple researchers present an experiment showing that multilingual self-supervised speech models (based on HuBERT) trained on English/French data underperform monolingual models on phonetic and linguistic measures when given a matched pretraining budget. Adding mechanisms that strengthen language discrimination during pretraining—an auxiliary language classifier or per-language k-means targets—reduces this gap: phone discrimination error drops from 11.6% to 10.4% (monolingual baseline 10.8%), lexical performance (sWUGGY) rises from 52.1% to 56.7% (monolingual 58.5%), and prosodic performance rises from 68.9% to 72.9% (monolingual 72.6%). The gains are largest when discrimination is introduced early in training, suggesting a causal role for language discrimination in reducing the cost of multilingual learning. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much does adding a language discrimination mechanism improve multilingual HuBERT speech models compared to monolingual baselines? In a controlled English/French HuBERT setup, adding an auxiliary language classifier or per-language k-means targets reduced phone discrimination error (phone-ABX) from 11.6% to 10.4%, versus 10.8% for monolingual models. Lexical performance (sWUGGY) rose from 52.1% to 56.7% (monolingual: 58.5%), and prosodic performance (ProsAudit) rose from 68.9% to 72.9% (monolingual: 72.6%). _Researchers comparing multilingual versus monolingual speech model tradeoffs can follow this kind of benchmark work via daily.dev._ ### When during HuBERT pretraining should language discrimination be introduced for the best linguistic performance gains? Introducing language discrimination in the first training iteration yields the strongest gains on most linguistic measures. Later or repeated interventions produce smaller improvements and increase language-wise segregation in the learned representations, indicating earlier intervention is more effective for closing the multilingual-monolingual performance gap. _Teams designing multilingual speech pretraining pipelines can track findings like this on daily.dev._ ## Similar posts on daily.dev - [Mix, Don’t Tune: Bilingual Pre-Training Outperforms Hyperparameter Search in Data-Constrained Settings](https://daily.dev/posts/mix-don-t-tune-bilingual-pre-training-outperforms-hyperparameter-search-in-data-constrained-settin-vgyf8wgth) · Apple Machine Learning Research · 0 upvotes · 0 comments - [Medium](https://daily.dev/posts/medium-vrsqfgio1) · Medium · 1 upvotes · 0 comments - [Google DeepMind Introduces ATLAS Scaling Laws for Multilingual Language Models](https://daily.dev/posts/google-deepmind-introduces-atlas-scaling-laws-for-multilingual-language-models-7e2kjgdrd) · InfoQ · 1 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#apple](https://daily.dev/tags/apple), [#speech-recognition](https://daily.dev/tags/speech-recognition) [View this post on daily.dev](https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Language Discrimination Improves Linguistic Learning in Multilingual Speech Models","url":"https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump"},"datePublished":"2026-10-02T14:48:34.806Z","dateModified":"2026-10-02T14:51:34.318Z","description":"Apple researchers present an experiment showing that multilingual self-supervised speech models (based on HuBERT) trained on English/French data underperform...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8dda20dea5f9a8a0e25a1b7bb36a1a20?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8dda20dea5f9a8a0e25a1b7bb36a1a20?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Apple Machine Learning Research","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Apple Machine Learning Research","logo":"https://media.daily.dev/image/upload/s--mD-1uSe0--/f_auto,q_auto/v1785654191/logos/mlapple","url":"https://daily.dev/sources/mlapple"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,apple,speech-recognition","timeRequired":"PT1M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Apple Machine Learning Research","item":"https://daily.dev/sources/mlapple"},{"@type":"ListItem","position":3,"name":"Language Discrimination Improves Linguistic Learning in Multilingual Speech Models"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/language-discrimination-improves-linguistic-learning-in-multilingual-speech-models-szjjq0ump#faq","mainEntity":[{"@type":"Question","name":"How much does adding a language discrimination mechanism improve multilingual HuBERT speech models compared to monolingual baselines?","acceptedAnswer":{"@type":"Answer","text":"In a controlled English/French HuBERT setup, adding an auxiliary language classifier or per-language k-means targets reduced phone discrimination error (phone-ABX) from 11.6% to 10.4%, versus 10.8% for monolingual models. Lexical performance (sWUGGY) rose from 52.1% to 56.7% (monolingual: 58.5%), and prosodic performance (ProsAudit) rose from 68.9% to 72.9% (monolingual: 72.6%). Researchers comparing multilingual versus monolingual speech model tradeoffs can follow this kind of benchmark work via daily.dev."}},{"@type":"Question","name":"When during HuBERT pretraining should language discrimination be introduced for the best linguistic performance gains?","acceptedAnswer":{"@type":"Answer","text":"Introducing language discrimination in the first training iteration yields the strongest gains on most linguistic measures. Later or repeated interventions produce smaller improvements and increase language-wise segregation in the learned representations, indicating earlier intervention is more effective for closing the multilingual-monolingual performance gap. Teams designing multilingual speech pretraining pipelines can track findings like this on daily.dev."}}]} ```