AI accessible to all, in every language

Google's language technologies now serve speakers of more than 300 languages, covering over 7 billion people — roughly 86% of the global population. The milestone highlights how far AI translation has come since Google Translate launched in 2006, but the company stresses that thousands of living languages remain poorly represented or entirely absent from the digital world. Its response is a broad push to make machines understand language the way humans actually use it.

From Text Transcripts to Native Audio Intelligence

Traditional speech recognition followed a rigid, multi-step pipeline: transcribe audio to text, process it, then synthesize it back into speech. That approach strips out tone, pacing, emotion, and context. Real speech is messy — people hesitate, overlap, and mix languages mid-sentence, as with Spanglish or Hinglish.

Google's answer is training models like Gemini to process raw audio directly, grasping both sound and intent. Gemini 3.5 Live Translate now delivers real-time spoken translation across 70 languages and more than 2,000 language pairs, capturing code-switching and emotional cues. Gemini 3.5 Transcribe, the company's most accurate speech-to-text model yet, converts raw audio into polished text even in noisy settings, and powers Rambler on Android Gboard, which removes filler words, corrects grammar, and allows voice-commanded edits and seamless language switching.

Ambitious research underpins this. The Universal Speech Model, trained on 12 million hours of audio, uses cross-lingual transfer learning to apply patterns from data-rich languages to under-resourced ones — a key step toward Google's goal of supporting the world's 1,000 most-spoken languages. The effort builds on 25 years of open research and more than 400 peer-reviewed speech papers.

Communities at the Heart of Language Data

Because the web overrepresents a handful of dominant languages, Google rethought data collection through grassroots partnerships. Three stand out:

• WAXAL (Wolof for "speaking"), built with Makerere University and Digital Umuganda, is a large-scale open speech dataset covering 27 Sub-Saharan African languages spoken by over 100 million people across more than 26 countries.
• Project Vaani, with the Indian Institute of Science and Bhashini, takes a region-anchored rather than language-anchored approach, collecting over 30,000 hours of speech across 109 languages from more than 155,000 speakers in India.
• The Amplify Initiative brought together more than 1,600 local experts and 20 universities across four continents — including Brazil's UFMG, India's IIT Kharagpur, and Uganda's Makerere University — contributing 15,000 multimodal data points.

A new interactive tool, Language Explorer, visualizes LinguaMeta, billed as the world's largest open-source language data repository, mapping more than 7,000 spoken, written, and signed languages. Google.org-backed efforts such as the Centre for Digital Language Inclusion and AI Singapore's Project Aquarium aim to turn this data into practical multilingual tools for farmers, healthcare workers, and teachers.

AI That Works Offline and on Basic Phones

More than 3 billion people still lack reliable internet access, so Google built TranslateGemma, a family of lightweight open translation models derived from Gemini and trained across 55 languages. Running efficiently on-device, it delivers quality translation without a cloud connection.

For the hundreds of millions still using feature phones, Google supports Viamo's "Ask Viamo Anything" (AVA), a voice AI assistant that brings Gemini to standard handsets. Piloted in Rwanda with existing interactive voice response users, AVA has already answered more than 2 million questions using Gemini.

Accessibility and Cultural Context

Conventional speech tools often fail people with non-standard speech, forcing users to adapt to the technology. Google's Sign Language-to-Text (SL2T) model, trained across 50+ sign languages, powers sign-to-text dictation in Gboard and Live Transcribe on Pixel 11, starting with American Sign Language to English — a first step toward serving the 70 million people worldwide who rely on sign language.

Cultural detail matters too. In New Zealand, Google worked with Māori language experts to improve place name pronunciation in Google Maps, ensuring text-to-speech models reflect local heritage rather than flattening it.

What Comes Next

Google says its language technologies now reach more than 5 billion people across nine platforms, including Search, Android, Chrome, YouTube, and Google Play. The company frames scale as only part of the mission: the deeper goal is systems that grasp context and respect culture. As it expands AI into broader societal applications, continued collaboration with local language communities — and progress toward the 1,000-language target — will be the metrics to watch.

Meta description: Google's language AI now supports 300+ languages for 86% of the world, with Gemini translation tools, open datasets, and offline models.

Tags: Google, AI translation, Gemini, speech recognition, low-resource languages

Featured image: Abstract visualization of interconnected sound waves and language symbols in Google brand colors, technology illustration style, no people or logos.