Google has launched Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, 2 new text-to-speech fashions in its Gemini Audio household. Google calls them its most expressive audio technology fashions but. Flash TTS targets inventive route and character voices. Flash-Lite TTS targets high-volume, cost-efficient manufacturing. Each let builders direct supply line by line utilizing pure language.
Is it deployable? Sure, each fashions are rolling out now via the Gemini API and Google AI Studio. Entry is API-only, with no open weights for self-hosting. Enterprise API entry through Gemini Enterprise is listed as coming quickly.
What Google Shipped
The discharge splits TTS into 2 tiers with shared route controls:
- Gemini 3.8 Flash TTS is constructed for deep inventive route and character design. Goal makes use of embrace gaming, immersive audiobooks, podcasts and interactive media. It affords granular management over appearing cues, pacing, dialect shifts and backchanneling.
- Gemini 3.8 Flash-Lite TTS is constructed for high-volume, cost-efficient scale. Google positions it for dubbing, audio content material creation and expressive voice brokers. It affords fine-grained management over tone, pacing and expressive nuance.
In AI Studio, the playground hyperlinks use the mannequin identifiers gemini-3.8-flash-tts and gemini-3.8-flash-lite-tts.
Voice Design From a Textual content Immediate
Earlier Gemini TTS provided 30 unique voices. The three.8 launch strikes to a a lot bigger voice system.
- Generative voice design: Flash TTS creates new voices from prompts describing position, accent and voice traits. This works throughout greater than 100 languages and dialects. Google’s demos embrace a Melbourne DJ, a monotone robotic and a Japanese dragon.
- Voice library: Builders get 2,000+ production-ready voices. Protection consists of regional varieties like Mexican Spanish, Quebec French and Scots English.
- Save and scale: Customized voices may be saved and reused, with minimal drift throughout initiatives.
- Voice remixing (coming quickly): Customers will alter a library voice’s timbre, pitch, tempo and accent via prompts.
Directing the Efficiency
Each fashions settle for stage instructions written within the script. Gemini may also steer supply from pure script cues.
- Lengthy-form technology: Voice high quality, pacing and timbre maintain throughout hours of steady audio.
- Native 2-speaker staging: A single script drives a multi-turn dialog with distinct, separated voices.
- Vocal bursts: Non-verbal cues like
,andadd conversational texture. - Backchanneling: Energetic-listening interjections like
|mhm|and|yeah|management response beats and comedic timing.
Voice Replication and Security Controls
Voice replication builds a constant vocal profile from a 30-second audio pattern. The pattern should be your voice or one you’ve gotten rights to make use of. Replication requires a verbal consent recording from the voice proprietor, matched in opposition to the reference speaker.
Each clip from Gemini Audio fashions carries a SynthID watermark. This imperceptible mark is embedded straight within the audio output. Replicated voices additionally carry C2PA content material credentials. Google factors to the Gemini 3.8 Audio model card for its broader security strategy.
Benchmark Outcomes
Google experiences these outcomes for the brand new fashions:
- Hume AI Voice Design Benchmark: Flash TTS ranks #1 general with a rating of 71.4, per Hume AI.
- Accent modeling: Flash TTS leads with a rating of 60.8.
- Hume AI General High quality Index: Flash TTS ranks #1 and Flash-Lite TTS ranks #2.
- Voice Arena blind desire: Each fashions take prime positions in Japanese, Brazilian Portuguese, Vietnamese, Trendy Customary Arabic, Mexican Spanish and Hindi.
/* step 1 personas */
var personas=[
{prompt:’A high-energy radio DJ from Melbourne, hyping up the next track.’,role:’Radio DJ’,accent:’Australian English’,ch:’High energy, fast pace’,w:{amp:.95,freq:4.5,jit:.35}},
{prompt:’A super-tinny, monotone robot reading out status updates.’,role:’Robot’,accent:’Synthetic, flat pitch’,ch:’Monotone, tinny timbre’,w:{amp:.4,freq:9,jit:.02}},
{prompt:’A dramatic Japanese dragon speaking from its mountain lair.’,role:’Fire-breathing dragon’,accent:’Japanese’,ch:’Deep, slow, dramatic’,w:{amp:.8,freq:1.4,jit:.12}}
];
var typer;
operate persona(i){
var p=personas[i];$$(‘#mtp-personas .chip’).forEach(operate(c,j){c.classList.toggle(‘on’,j===i)});
$(‘#mtp-role’).textContent=p.position;$(‘#mtp-accent’).textContent=p.accent;$(‘#mtp-char’).textContent=p.ch;
w1.set(p.w);clearInterval(typer);var el=$(‘#mtp-prompt’),n=0;
if(cut back){el.textContent=p.immediate;return}
typer=setInterval(operate(){n++;el.innerHTML=p.immediate.slice(0,n).exchange(/,'<‘)+’‘;if(n>=p.immediate.size)clearInterval(typer)},22);
}
$$(‘#mtp-personas .chip’).forEach(operate(c){c.addEventListener(‘click on’,operate(){persona(+c.dataset.p)})});
persona(0);
/* step 2 script */
var baseWords=[‘I’,’checked’,’the’,’numbers’,’twice,’,’and’,’the’,’launch’,’still’,’goes’,’out’,’on’,’Friday.’];
var cues=[];
operate renderScript(){
var html=””,pos={};cues.forEach(operate(c)[]).push(c));
for(var i=0;i’+c.label.exchange(/,'<‘).exchange(/>/g,’>’)+’ ‘});
if(i
];
var taking part in=false;
$(‘#mtp-play’).addEventListener(‘click on’,operate(){
if(taking part in)return;taking part in=true;var A=$(‘#mtp-la’),B=$(‘#mtp-lb’);
$$(‘#mtp-g38tts .seg’).forEach(operate(s){s.take away()});
scene.forEach(operate(sg,i){
var tr=sg.lane===’a’?A:sg.lane===’b’?B:(sg.tr===’a’?A:B);
var d=doc.createElement(‘div’);d.className=”seg “+sg.lane;d.fashion.left=sg.s+’%’;d.fashion.width=(sg.e-sg.s)+’%’;d.textContent=sg.t;tr.appendChild(d);
setTimeout(operate(){d.classList.add(‘present’);$(‘#mtp-cap’).textContent=sg.cap},cut back?0:i*1100+100);
});
$$(‘#mtp-g38tts .playhead’).forEach(operate(p){p.fashion.opacity=1;p.fashion.transition=’none’;p.fashion.left=”0″;void p.offsetWidth;p.fashion.transition=cut back?’none’:’left 5.6s linear’;p.fashion.left=”100%”});
setTimeout(operate(){taking part in=false;$(‘#mtp-play’).textContent=”Replay scene”;$$(‘#mtp-g38tts .playhead’).forEach(operate(p){p.fashion.opacity=0})},cut back?50:5800);
});
/* step 4 examine */
var fashions=[
[[‘Built for’,’Deep creative direction and character design’],[‘Best fit’,’Gaming, immersive audiobooks, podcasts, interactive media’],[‘Control’,’Acting cues, pacing, dialect shifts, backchanneling’],[‘Headline feature’,’Generative voice design from scratch with prompts’],[‘Hume AI Overall Quality Index’,’#1′],[‘Available today’,’Gemini API, Google AI Studio, Gemini Notebook’]],
[[‘Built for’,’High-volume, cost-efficient scale’],[‘Best fit’,’Dubbing, audio content creation, expressive voice agents’],[‘Control’,’Tone, pacing, expressive nuance’],[‘Headline feature’,’Expressive output tuned for volume and cost’],[‘Hume AI Overall Quality Index’,’#2′],[‘Available today’,’Gemini API, Google AI Studio, Google Vids’]]
];
operate mannequin(i){
$$(‘.toggle button’).forEach(operate(b,j){b.classList.toggle(‘on’,j===i)});
$(‘#mtp-cmp’).innerHTML=fashions[i].map(operate(r,j){return ”}).be part of(”);sendH();
}
$$(‘.toggle button’).forEach(operate(b){b.addEventListener(‘click on’,operate(){mannequin(+b.dataset.m)})});
mannequin(0);
/* step 5 scores */
operate runScores(){
$$(‘#mtp-g38tts .rating .n’).forEach(operate(el){
var to=parseFloat(el.dataset.to),st=null;
if(cut back){el.textContent=to.toFixed(1);return}
operate f(ts){if(!st)st=ts;var okay=Math.min(1,(ts-st)/1100);el.textContent=(to*(1-Math.pow(1-k,3))).toFixed(1);if(okay
“>
Key Takeaways
- Google launched Gemini 3.8 Flash TTS for inventive work and Flash-Lite TTS for scale.
- Flash TTS designs new voices from prompts throughout 100+ languages and dialects.
- Builders get 2,000+ manufacturing voices, up from 30 originals.
- Voice replication wants a 30-second pattern plus an identical consent recording.
- Flash TTS ranks #1 on Hume AI’s Voice Design Benchmark with 71.4.
FAQ
- What’s Gemini 3.8 Flash TTS? It’s Google’s text-to-speech mannequin for inventive voice design and line-by-line efficiency route. It’s obtainable via the Gemini API and Google AI Studio.
- How is Flash-Lite TTS totally different? Flash-Lite TTS is optimized for high-volume, cost-efficient workloads like dubbing and voice brokers. It ranks #2 on Hume AI’s General High quality Index.
- Can I clone my very own voice? Sure, with a 30-second pattern and a verbal consent recording. It’s unavailable in AI Studio in a number of areas, together with the UK, EEA and India.
Take a look at the Technical Blog. All credit score goes to the researcher of this challenge. Additionally, be at liberty to comply with us on Twitter and don’t neglect to affix our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
Have to companion with us for selling your GitHub Repo OR Hugging Face Web page OR Product Launch OR Webinar and so on.? Connect with us
Asif Razzaq is the CEO of Marktechpost AI Media Inc.. As a visionary entrepreneur and engineer, Asif is dedicated to harnessing the potential of Synthetic Intelligence for social good. His most up-to-date endeavor is the launch of an Synthetic Intelligence Media Platform, Marktechpost, which stands out for its in-depth protection of machine studying and deep studying information that’s each technically sound and simply comprehensible by a large viewers. The platform boasts of over 2 million month-to-month views, illustrating its reputation amongst audiences.

