Google Rolls Out 66 Gemini Regional Voices, Including UK Geordie, Scouse and Brummie Accents
1 day ago / Read about 27 minute
Source:TechTimes

Vincent Feuray/gettyimages.com

Google began rolling out 66 regional voices in the Gemini app on Android and iOS on Monday, November 16, adding UK dialects such as Geordie, Scouse and Brummie so Gemini can answer aloud in speech that matches where you live. The launch also touches a security question: Abertay University researchers have found that listeners often assume an AI voice is a real person, particularly when it speaks in a regional dialect, though their studies tested other voices, not Gemini's.

Voice is now how most people use Gemini. Google reported in August that 63% of Gemini users interact with the assistant by voice, alongside the news that the app passed one billion monthly users. The new setting changes what Gemini sounds like in Gemini Live, in language practice, and when it reads a reply aloud through the Listen icon.

Google Names Seven Languages and Locales, Not Individual Dialects

Google's announcement lists U.S. English, U.K. English, Indian English, Hindi, Brazilian Portuguese, Indonesian and Korean, with voices for each. The post names no individual dialects and gives no count of how many of the 66 voices go to each language.

Independent outlets fill part of the gap, and they disagree on the list. 9to5Google names eight UK options (Geordie, Scouse, Brummie, Scottish, Welsh, Northern Irish, Modern RP and Multicultural London English) and says there are more than ten. Android Headlines names seven and also reports more than ten. The Independent names five: Geordie, Scouse, Brummie, Scottish and Welsh. None of the lists accounts for the "more than ten" count, so the full UK lineup is unconfirmed.

Google says each voice is tuned for natural cadence and everyday clarity "rather than exaggerated accents." The Independent reports Google built the voices with native speakers, and Android Headlines attributes that to Neel Joshi, director of product management for Gemini Live; Google's own post does not say so. Google shared plans at I/O in May to bring local dialects to Gemini, and Business Today reports a live demo of Gemini Live speaking Haryanvi, spoken in Haryana, India.

How Do You Switch Gemini's Voice to Geordie or Scouse?

Open the three-dot menu inside Gemini Live, or go to "Gemini's voice" in the Gemini app settings. You can preview each voice before choosing, and the choice applies across the app, including Listen playback on chat replies.

Two quirks complicate the choice. The picker does not label voices by city or region, and Gemini is not programmed to identify its accent if you ask, so you have to listen to find your Geordie. Google suggests prompts about local food, culture and sport to bring out each voice's rhythm, The Independent reports.

Gemini Live Offered One British Voice in 2024, Then Accents on Request in 2025

When Google added ten English voices to Gemini Live in September 2024, one of them, Capella, carried a British accent. On November 12, 2025, Google said Gemini Live could speak in accents on request, with a Cockney British accent for a Sunday roast recipe as one example.

The November 16 rollout makes accent a selectable voice choice, and 9to5Google counts more than ten UK English options against that single British voice in 2024.

Developer Docs Steer Accent Through Chosen Voice, Not Style Prompt

Google's developer documentation for Gemini text-to-speech advises against permanent accent changes in the style field. It recommends choosing a regional voice from an Extended Voice Library that can be filtered by accent, or creating one with Voice design, and it lists 30 prebuilt studio voices. The page calls long style blocks the most common cause of voice drift.

That design is consistent with Google shipping 66 distinct, selectable voices instead of one accent dial. It is context, not confirmation: Google has not said how the app's regional voices were built or which model produces them, and the developer documentation describes a separate product, the Gemini API.

The audio model behind Gemini Live is documented separately. Gemini 3.1 Flash Live, which Google launched on March 26, 2026 for Gemini Live and Search Live, works audio-to-audio: it ingests raw speech and produces audio output end to end rather than converting speech to text and back. Google says all audio the model generates is watermarked with SynthID, an imperceptible marker meant to help detect AI-generated content.

Accent stability is the open engineering question. On March 30, 2026, 9to5Google reported Gemini Live voices had changed in cadence and tone, sometimes sliding into Australian or standard American accents mid-conversation, and said the British Capella voice had changed or broken down significantly. For the new rollout, 9to5Google writes that the update might help reduce the drift but that its author has not yet been able to test whether it is resolved. Android Headlines writes that the update should also fix British voices slipping into American or Australian hybrids, an expectation rather than a tested result. No listening test of the new voices appears in the sources reviewed.

One Commercial Speech Recognizer Averaged 32% Word Errors on Newcastle English

The new voices change what Gemini says back, not how well it hears you, and Google's post says nothing about whether recognition of these accents improved. Dana Serditova, Kevin Tang and Jochen Steffens tested a commercial speech recognizer, Rev AI, on about 72 hours of Tyneside speech from 160 speakers in the Diachronic Electronic Corpus of Tyneside English, and it averaged a 31.95% word error rate across the full dataset. The paper's error analysis pointed to Newcastle features such as glottalized /t/, which turned "patent" into "paint," plus g-dropping and local words.

The pattern is not new. A 2020 study in PNAS led by Allison Koenecke found five commercial recognition systems averaged a 0.35 word error rate for Black speakers against 0.19 for white speakers.

Abertay Research Links Regional Dialects to AI Voices Passing as Human

Dr. Neil Kirk of Abertay University led a 2025 study in which 100 adults in Scotland and 100 in southern England judged short phrases recorded in Scottish Standard English and Dundonian Scots, then processed with AI software. Scottish participants correctly identified genuine voices 57% of the time and English participants 52%, and Scottish participants were more likely to perceive AI voices speaking Dundonian Scots as human, though the university published no figures for that comparison. Kirk warned that AI-augmented voices "could be used in scams like phishing or fraud."

A follow-up study in the Journal of Cybersecurity, announced by Abertay on March 2, 2026, ran two experiments with 300 Scottish participants. Messages explaining AI capabilities told people that AI can convincingly copy Scottish speech and significantly reduced the tendency to classify AI voices as human, while warnings that only described scam risks had little effect unless paired with that information. Listeners' ability to tell human voices from AI voices did not improve. Abertay defines the MINDSET bias as Minority, Indigenous, Non-standard, and Dialect-Shaped Expectations of Technology: an assumption that voice systems cannot handle local or regional speech, which makes people more likely to trust AI voices that use dialects. Abertay cites UK Finance data putting the average loss from a deepfake scam call at £595 (approximately $786 USD), with some cases above £13,000 (approximately $17,181 USD), and attributes the figures to UK Finance's Annual Fraud Report 2025.

Both Abertay studies tested other voices, not Gemini's. None of the sources reviewed shows Gemini's regional voices being used in a scam, but The Independent cited and Android Headlines cited the Abertay warning in their launch coverage. Kirk's recommended defense is public awareness campaigns or security prompts from services such as online banking, even at the cost of some false alarms.

Google Has Not Said Which Model Powers Voices, or Per-Language Counts

Google's November 16 post does not name the model behind the 66 voices. Google says audio from Gemini 3.1 Flash Live is SynthID-watermarked, so whether these voices carry the watermark is unconfirmed.

The post gives no per-language voice counts, no full rollout schedule and no per-region availability. No independent listening test of the new voices is available in the sources reviewed, so whether "Scouse" holds up across a long conversation is untested.

A 2025 Northeastern University study of two other synthetic voice services, Speechify and ElevenLabs, found participants felt American and British accents dominate synthetic voices. Some said their voice clones had over-emphasized accents, and one participant said the only Indian-accent option made them feel mocked. Google says its aim is to avoid exaggeration; whether Gemini's voices achieve that is unverified.

For Gemini users, the practical step is to preview several voices, then test one in a long conversation to see whether the accent holds, since drift has been documented in Gemini Live. For everyone, Abertay's research carries the larger point: a voice that sounds like your neighborhood is not evidence that a person is speaking.

Exchange rate as of November 16, 2026; conversions are approximate.

Frequently Asked Questions

How do I change Gemini's voice to a regional accent?

Open the three-dot menu in Gemini Live or choose "Gemini's voice" in the app settings, preview the options, and confirm. The voice then applies across Gemini Live, language practice and Listen playback.

Can AI voices that sound local be used in scams?

Abertay researchers warned that AI-augmented voices could be used in phishing or fraud, and their 2026 work finds people are more likely to trust AI voices that use dialects. No source reviewed ties Gemini's regional voices to any scam. In the United States, the Federal Communications Commission ruled on February 8, 2024, that calls made with AI-generated voices count as "artificial" under the Telephone Consumer Protection Act, which brings them under prior-consent rules for robocalls.

Will Gemini understand my accent better now?

Google's announcement covers how Gemini sounds, not how well it recognizes speech, and it does not report any recognition improvement. Studies of other commercial recognizers have found higher error rates for regional speech, including Newcastle English.

Are the regional voices available on every phone yet?

Google says it is rolling the voices out on Android and iOS and gives no full rollout schedule or per-region availability, so a voice may take time to appear in your app.