Google Drops WAXAL + Opt‑Out
Google released WAXAL, a multilingual African speech dataset for ASR/TTS training, and separately said websites can opt out of Google's generative AI features to address UK competition concerns—two moves on data access and control. The dataset expands global ASR resources while the opt‑out signals evolving web data governance. (marktechpost.com) (reuters.com)
WAXAL covers 21 Sub‑Saharan African languages and represents speech from more than 100 million speakers, with an ASR component of roughly 1,250 hours and a TTS component of over 180 hours. (arxiv.org) The dataset is released under a permissive CC‑BY‑4.0 license and published on Hugging Face as google/WaxalNLP, with collection and curation carried out in partnership with four African academic and community organizations. (arxiv.org) WAXAL’s ASR material was elicited via image‑prompted, natural speech recordings, while the TTS portion comprises studio‑quality, single‑speaker recordings reading phonetically balanced scripts. (arxiv.org) The arXiv paper documents QC, annotation pipelines, and explicit attention to tonal variability and code‑switching to help benchmark models for low‑resource linguistic phenomena. (arxiv.org) Google told the UK regulator it is developing new Search controls that would let websites opt out specifically of generative‑AI features such as AI Overviews and AI Mode. (blog.google) The UK Competition and Markets Authority designated Google as having “strategic market status” in October 2025 and has proposed conduct requirements that would let publishers opt out of AI summaries without losing general search visibility. (gov.uk) In its formal response on the CMA consultation, Google warned some proposed conduct requirements would have “disproportionate and detrimental consequences” and proposed alternatives such as a “less intrusive” device‑level switch to change default search engines. (uk.finance.yahoo.com)
