Dan Steinman retweeted
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
Here's a sneak peak at the Jaxcore Control NRF54 - this uses the upcoming Raytac AN54LM-P20B. It has 512 KB ram, 2MB NVM flash, SD slot on the bottom.
Like my ESP32-S31 design, this is a minimalist breakout with no onboard power supply. Bring your regulator, or stay tuned for my Buck/Boost/PMIC modules. The 3.0V PMIC I'm working on should make this board a great option for battery powered projects.
#ESP32 #KiCad #NRF54 #Bluetooth
ESP32-S31 also
乐鑫(Espressif)官方开源了一款非常亮眼的嵌入式利器:基于 Rust 编写的 esp-emulator,无需任何实体开发板,就能在本地终端乃至浏览器中直接模拟运行 ESP32 系列 SoC 固件。
这次更新与功能覆盖相当全面,直击嵌入式开发者和 CI 自动化测试的痛点:
广泛的芯片支持:全面覆盖 RISC-V 架构的 ESP32-C3、C5、C6、H2、P4、S31,以及 Xtensa LX7 双核的 ESP32-S3。不仅模拟 CPU 指令集与 FPU,还还原了 PMP/PMA 权限控制、MMU 与加密加速器。
完整的无线协议栈仿真:
Wi-Fi:内置 Soft AP,支持 WPA2/WPA3 握手、企业级 802.1X 以及 DHCP 服务。
用户态免配置网络(User-mode NAT):无需 sudo 或 TAP 网卡,开箱即用实现 TCP/UDP 转发;更自带 mDNS 记录重写,支持无缝调试 Matter / HomeKit 配网。
BLE & Thread:BLE 支持配合 Google Bumble 进行纯软件仿真;ESP32-C6 和 H2 支持通过 UDP 桥接两台仿真器,组成两节点的 802.15.4 Thread 真实 Mesh 网络。
无缝嵌入现有开发工作流:
支持 --uart-tcp 桥接,esptool.py、espefuse.py 和 idf.py flash 可以直接通过 socket:// 烧录固件与 eFuse。
内置 GDB 远程调试存根(--gdb),并提供 --trace 输出 Perfetto/Chrome 跟踪文件,无需在固件内插桩即可精确观察 FreeRTOS 任务调度与 CPU 耗时。
零安装体验:官方构建了 WebAssembly 版本并直接部署在 GitHub Pages,只需把编译合并好的 merged-binary.bin 拖拽到网页中,就能实时查看控制台输出、GPIO 状态、寄存器与内存。
单行脚本一键安装体验: curl -fsSL raw.githubusercontent.com/es… | sh
做固件 CI/CD、网络协议栈验证或手头没有板子时的救急调试,推荐务必收藏体验!
#ESP32 #Espressif #RustLang #IoT #Embedded #OpenSource
Yup they did it
Replying to @Polymarket
Sounds like when Knight Rider got K.I.T.T. upgraded with Turbo Boost mode so it could jump over cars or obstacles. Jumping over traffic is gonna be sick.
I'm working on some web-based audio software and it accidentally turned into a DAW. Sol 6.1 knows a lot more about music than I do. #DAW
Dan Steinman retweeted
NVIDIA really removed the biggest friction point in Voice AI 🤯
They open-sourced PersonaPlex-7B, a full-duplex conversational model that can listen and speak at the same time.
Instead of waiting for you to finish talking, it uses a dual-stream architecture to process and respond in real-time.
100% Open-Source and Free.
There's really no good way to route a USB-C port is there? I hate having to redo this each time. This is a specific task where I think AI routing could be useful.
I've been playing with Qwen3-TTS 1.7B today.
I found out that this Apple native MLX version runs roughly 10X faster than the PyTorch version on Apple silicon. Now we're talking.
github.com/Blaizzy/mlx-audio
I'm delighted at my ability to give Sol 6 problems that require over an hour for it to solve. Literally a month or more of human coding work is being done while I drink my coffee and read the news.
Yeah I've never been this excited. AI is unlocking so much potential for things I would have never done by hand.
I used Sol-6 to do a big port and massive feature upgrade of an existing Javscript project to C -> WASM -> Arduino -> Python -> Node -> Web (back to JS for web browsers).
Then I ported the base C / WASM implementations to a common Rust base implementation. Then upgraded the Arduino, Python, Node, Web codebases to use the new Rust pipeline.
No sane programmer would ever even attempt to do all this by hand. And I didn't have to write a single line of code and it's working beautifully.
AI coding is now way, way better than I ever imagined. After seeing the potential for myself I have a new perspective on what kinds of software can be built now.
Cuatom silicon is really the next frontier of electronics.
Instead of designing just a PCB, ChatGPT can also design a custom integrated integrated circuit (ASIC) that you can send off to get fabricated into silicon. Services like GlobalFoundaries or CMC Microsystems then sends your design to TSMC to get fabricated along with lots of others in big batch production runs.
You Can Make a Microprocessor That’s All Your Own ift.tt/hyZmL7T
AI reset anxiety is real
1 weekly reset remaining
6 days until monthly reset
and a lot more features are still needed
#nPM1300 is terrific, has lipo charging and 2 buck regulators in one IC. I have a maker oriented PMIC module in the works using this. It's absolutely perfect match for powering #NRF54 battery powered #BLE projects
.@PlayerData's Edge Air tracker utilizes #nPM1300 to achieve up to 12-hour battery life and #nRF52840 for core processing and #Bluetooth LE connectivity.
“Power management is the first step towards having a good product, and by incorporating the highly configurable nPM1300 in our solution, we have addressed this key element of any sports data tracker while achieving an impressive form factor to help target the mass market. Prior to using the nPM1300, we relied on four discrete ICs to do the same job."
Why isn't there an AI chat bot that works in reverse? Instead of the human asking questions and the AI answering, the AI should ask humans the questions it doesn't know the answer to or is curious about.
This seems like one of the missing pieces to AGI.
On a few occasions I've caught ChatGPT swapping in an emdash instead of a regular dash when showing numbers. Like c'mon
I'm starting to hate this screen. What is this doing and why does it take so long? What are the consequences of killing the process? #Kicad



















