Assertion Supported AI assessment confidence: 95% certainty 4/5 debate potential 1/5

Mallick: Gemini Live API allows custom VAD tuning and third-party integration

Shrestha Basu Mallick · [AIEWF Preview] Gemini in 2025 and Realtime Voice AI · Jun 2, 2025 · at 17:30

Shrestha Basu Mallick, product leader at Google DeepMind, discusses voice infrastructure and API controls available for real-time Gemini voice applications.

0:00 / 0:19exact quote · 19.9s

▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark

“Now developers can actually tune the sensitivity on our voice activity detection model as well as, you know, how much of the prefix, like how much of a time duration at the beginning, at the start or stop of saying things. And we also have a mode where now you can, where you can disable our voice activity detection and bring your own.”

quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →

The claim, plainly

Google allows developers using the Gemini Live API to tune server-side Voice Activity Detection sensitivity and prefix duration, or disable it completely to bring their own VAD.

a quote alone can be ambiguous, so this restates it with the references and context of the conversation around it resolved. AI-written; the tape below the player is the source.

Assessment

Official Google Gemini Live API documentation and SDK specifications confirm all aspects of the claim. Developers can configure server-side Voice Activity Detection (VAD) via `RealtimeInputConfig.automaticActivityDetection`, which exposes `startOfSpeechSensitivity`, `endOfSpeechSensitivity`, `prefixPaddingMs`, and `silenceDurationMs`. In addition, setting `disabled = true` disables the server-side automatic VAD, enabling developers to implement custom or client-side VAD by sending explicit `activityStart` and `activityEnd` control messages.

Google Gemini Live API documentation confirms `RealtimeInputConfig.automaticActivityDetection` exposes parameters `startOfSpeechSensitivity`, `endOfSpeechSensitivity`, and `prefixPaddingMs`.
Google documentation specifies that setting `automaticActivityDetection.disabled = true` disables server-side VAD, allowing developers to bring their own VAD and signal turns via `activityStart` and `activityEnd` messages.
The Google Gen AI SDK supports `types.AutomaticActivityDetection` with fields `disabled`, `start_of_speech_sensitivity`, `end_of_speech_sensitivity`, and `prefix_padding_ms`.

Checked against

[1] google.com [2] google.dev [3] google.dev [4] google.dev [5] google.dev [6] google.dev [7] github.io [8] dev.to

Gemini Live API Voice Activity Detection VAD disable bring your own VAD documentation site:ai.google.dev "automaticActivityDetection" Gemini Multimodal Live API voice activity detection VAD sensitivity prefix duration disable site:ai.google.dev "prefixPaddingMs" OR "silenceDurationMs" OR "startSensitivity" "RealtimeInputConfig" "AutomaticActivityDetection" "gemini" "automaticActivityDetection" "prefix_padding_ms" OR "prefixPadding" "gemini"

assessed as of 2026-09-02

This assessment is an AI reading of a public statement, made against the cited references with the confidence shown. It can be wrong. The quote, the tape and the references are all on this page; if something is off, use report an error below. full disclaimer →

Mentioned

Gemini Live API Google DeepMind Voice Activity Detection

⚑ Spot an error? Report it

More from Shrestha Basu Mallick

Prediction Not checkable as stated

Basu Mallick: Most voice use cases will transition to audio-to-audio models

“I do think perhaps eventually For most use cases, as these audio to audio architecture models get better, a lot of use cases will probably transition to that.”

Shrestha Basu Mallick Jun 2, 2025 ▶ 11:29 [AIEWF Preview] Gemini in 2025 and Realtime Voice AI

Assertion Contradicted

Mallick: Gemini recognizes distinct voices as an unsupported emergent behavior

“This is not officially supported yet. The model just does it.”

Shrestha Basu Mallick Jun 2, 2025 ▶ 21:31 [AIEWF Preview] Gemini in 2025 and Realtime Voice AI

Assertion Supported

Mallick: Google was first to market with live video API input

“So we were actually the first to market with also video input”

Shrestha Basu Mallick Jun 2, 2025 ▶ 7:53 [AIEWF Preview] Gemini in 2025 and Realtime Voice AI

Assertion Supported

Basu Mallick: Google was first to introduce tool chaining for live APIs

“Again, we were very proud because we introduced tool chaining first, so you could change search and code execution to all kinds of analysis.”

Shrestha Basu Mallick Jun 2, 2025 ▶ 8:33 [AIEWF Preview] Gemini in 2025 and Realtime Voice AI

Prediction Not checkable as stated

Mallick: Google's URL Context tool will enable developer research agents

“And I think that this will unlock new use cases. Like if people want to build their own version of a research agent, which is something developers ask us for a lot.”

Shrestha Basu Mallick Jun 2, 2025 ▶ 3:43 [AIEWF Preview] Gemini in 2025 and Realtime Voice AI

Assertion Not checkable as stated

Basu Mallick: Transcription was one of Gemini Live API's biggest early use cases

“Transcription actually, even before we released native audio, now of course you get text and audio interleaved in the output, but transcription used to be one of the biggest use cases we had on the live API.”

Shrestha Basu Mallick Jun 2, 2025 ▶ 7:25 [AIEWF Preview] Gemini in 2025 and Realtime Voice AI

Mallick: Gemini Live API allows custom VAD tuning and third-party integration

More from Shrestha Basu Mallick

Turn any episode into a week of clips.