Google keeps refining Gemini Live. The conversational voice feature first launched with promise of natural back-and-forth exchanges. Now an APK teardown shows the company testing another interface update. This one brings a fresh waveform animation and a dedicated speaker control.
The redesign surfaces in Google app version 17.65.17. Android Authority spotted it first. Testers activated the hidden code to reveal a work-in-progress screen. The bottom area now hosts a redesigned waveform. It pulses differently from previous versions. Up top, near the right corner, sits a speaker icon. Tap it and Gemini should stop talking. No need to adjust phone volume or hunt through settings.
But this early build feels stripped down. Video sharing and screen sharing buttons are absent. The prototype doesn’t respond to voice queries yet. Still, the direction looks clear. Google wants cleaner controls for voice output. Users often mute the AI mid-conversation. A direct button makes that simpler. One less friction point.
This change fits a longer pattern. Over the past year Google has repeatedly altered how Gemini Live appears. Earlier tests ditched the full-screen takeover. The mode now runs inside the main Gemini chat view. A pill-shaped container holds the waveform. Camera and mic controls flank it. Responses appear on screen as the conversation flows. Android Authority first detailed those shifts in April.
And the updates haven’t stopped. Neural Expressive styling arrived later. It added glowing effects, colorful dots, and softer shapes to overlays and Live screens. Floating widgets became circular. Icons slimmed down. The visual language grew more dynamic. Yet functionality kept expanding too.
Google rolled out Guided Vision in early October. The feature turns Gemini Live into a conversational visual assistant for blind and low-vision users. Share your camera feed. Gemini describes surroundings, offers proactive cues, and answers follow-up questions. It was built with input from the accessibility community and Aira. Real-time prompts help center the camera. “Pan slowly to the right” or “tilt downward” come naturally through voice. The Google Blog outlined the launch on October 1.
Model upgrades arrived around the same period. Gemini 3.8 Live and its Extended Thinking variant boosted reasoning speed and tool use. Background task execution no longer breaks conversation flow. The system handles 97 languages with automatic detection. Live Avatar adds synchronized video personas for enterprise deployments. These capabilities, announced in September by Google DeepMind, raise the bar for what a voice interface must support.
So the latest UI test arrives against this backdrop. The waveform change isn’t cosmetic alone. It pairs with a control that directly addresses a common pain. Lowering system volume to silence Gemini disrupts other audio. A dedicated mute fixes that. Early code suggests the speaker button targets voice output specifically.
Observers note the build remains incomplete. Missing buttons could return. Future versions might layer in the neural styling seen elsewhere. Integration with recent model improvements seems likely. Users already expect Live to handle camera input, screen context, and complex tasks without dropping context.
Google has stayed quiet on timing. No official comment accompanies the teardown. History shows these code discoveries often precede releases by weeks or months. The company continues to iterate quickly. Previous redesigns reached beta testers within months of first leaks.
Competitive pressure plays a role. OpenAI’s voice mode in ChatGPT and other assistants push similar natural dialogue. Google counters with deeper Android integration, camera access, and now accessibility tools like Guided Vision. The speaker icon fits this practical focus. Make voice mode usable in more situations. Reduce interruptions to daily phone use.
Look closer at the waveform itself. Earlier designs used simpler pulsing bars. The new animation appears more fluid. It may better reflect speech patterns or model activity. Exact visuals remain limited to demo videos in the teardown report. They show movement but no live responses. Still, the shift signals attention to detail in an area users stare at during every conversation.
Broader Gemini app changes provide context. Sidebar updates, Discover tabs, and thinking response sheets have appeared in other teardowns. The Live interface evolves alongside them. Nothing stays static. Each adjustment aims to make the AI feel less like a separate app and more like an always-available partner.
The speaker control could prove especially valuable during longer sessions. Planning a trip, debugging code, or simply chatting. Quick muting without volume tweaks keeps the experience smooth. Pair it with Guided Vision for users who rely on voice descriptions of the physical world. The combination points toward more inclusive, less intrusive AI.
Of course, teardowns come with disclaimers. Features spotted in code don’t always ship. Elements can change or disappear entirely. This redesign might look different by release. Buttons could reappear. Animation styles may receive further polish. Yet the core ideas — updated visual feedback and independent voice controls — address real user feedback.
Industry watchers on X reacted quickly to the Android Authority report. Many highlighted the mute button as overdue. Others speculated on ties to upcoming model features. The conversation reflects growing maturity around voice AI. Early excitement about natural talk has given way to demands for practical controls and accessibility.
Google’s pace suggests more updates soon. The company already ships new voices, memory access, and app integrations for Live. Visual understanding keeps improving. Each layer adds complexity to the interface. Clean, intuitive design becomes even more necessary.
This latest glimpse offers a narrow window into that work. A new waveform. A speaker icon. Small pieces. But they hint at larger goals. Make Gemini Live the default way many interact with AI on Android. Fluid enough for everyday use. Powerful enough for demanding tasks. And controllable without hassle.
Expect further refinements. The pattern holds. Code appears. Testers enable it. Feedback rolls in. Polish follows. Then rollout. For now, the speaker button stands out. Simple. Direct. And potentially quite useful.