🧰 Builders building with Gemini Audio
Read the Portuguese translation →
Author: Google for Developers · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: 🧰 Builders building with Gemini Audio · Channel: Google for Developers · 00:34 · Watch on YouTube
Topics: Enterprise AI applications, AI evaluation and benchmarks, AI content generation, AI models
Summary
Section titled “Summary”Builders discuss experimenting with Lyria as a partner in music creation and connecting a Reachy Mini robot to Gemini through WebSockets. They describe using voice models to help someone with visual impairment understand what is in front of her, and note that Gemma can be useful where Wi-Fi is unavailable. They also praise the responsiveness of live audio, Gemini’s Extended Thinking feature, and performance on the MAviS Bench speech-to-speech benchmark.
Key points
Section titled “Key points”- One speaker sees Lyria as a partner for music creation.
- Reachy Mini has a Gemini client that connects to Gemini in the cloud through WebSockets.
- Gemini Live or a Gemma model could use voice to help a person with visual impairment understand their surroundings.
- Gemma models may be useful in environments without Wi-Fi.
- Live audio is praised for fast responses and gathering context in real time.
- Extended Thinking can run a long process in the background while a separate conversation continues.
- MAviS Bench evaluated Gemini alongside other speech-to-speech models, and the speaker says Gemini performed very well.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:02 | Lyria and creative collaboration | A speaker describes Lyria as a partner for music creation. |
| 00:02 | Gemini, Reachy Mini, and voice accessibility | The speakers discuss a cloud-connected Reachy Mini and voice-based assistance for someone with visual impairment. |
| 00:34 | Live audio, Extended Thinking, and benchmarking | The speakers praise live audio and Extended Thinking, and describe Gemini’s strong results on MAviS Bench. |
Quotes
Section titled “Quotes”I’d like to think it more as a partner. for my music creation. — 00:02
a Gemma model to be able to use voice to actually see what’s in front of her. — 00:02
an agent can respond really quickly it gathers the context in real time. — 00:34
you guys are absolutely killing it. — 00:34
Concepts and entities
Section titled “Concepts and entities”Lyria, Gemini 3.8, Reachy Mini, WebSockets, Gemini Live, Gemma, Extended Thinking, MAviS Bench
Full transcript (original language)
Section titled “Full transcript (original language)”00:02 How have you been experimenting with the Lyria models? - I’m a big fan of them. I’d like to think it more as a partner. for my music creation. - This is Gemini 3.8, and Reachy Mini has a Gemini 3.8 client which is connecting through WebSockets to Gemini 3.8 running in the cloud. - I’m working with somebody right now that has some visual impairment. I was thinking of using Gemini Live or a Gemma model to be able to use voice to actually see what’s in front of her. - We discussed Gemma models. They are perfect because
00:34 in some environments you might not have a Wi-Fi connection. - We’re loving how live audio is feeling now. It’s, you know, an agent can respond really quickly it gathers the context in real time. - I love your guys’ Extended Thinking feature. That’s very cool that you can have a long running process go in the background while you have conversation that’s not related to the async tool call that’s going on. We recently released a benchmark called MAviS Bench where we benchmarked you guys along with a couple of other speech-to-speech models, and you guys are absolutely killing it.