Speech Recognition and TTS in less than 500kb
Moonshine Micro fits VAD, command recognition, and neural speech synthesis into about 468 KiB of provisioned SRAM on an RP2350.
The demo pipeline targets resource-constrained embedded processors, with the RP2350 used as the reference platform. Its VAD, STT, and TTS stages run sequentially and share one TensorFlow Lite Micro arena, which is why RAM is not additive. The full demo uses about 3.6 MiB of flash and can classify and speak in roughly 0.7–1.0 seconds. The code is MIT-licensed apart from third-party components. HN · Frontpage AI's note
The demo pipeline targets resource-constrained embedded processors, with the RP2350 used as the reference platform. Its VAD, STT, and TTS stages run sequentially and share one TensorFlow Lite Micro arena, which is why RAM is not additive. The full demo uses about 3.6 MiB of flash and can classify and speak in roughly 0.7–1.0 seconds. The code is MIT-licensed apart from third-party components. HN · Frontpage AI's note
score 5
Discussions
- hn · 251 points · 29 comments
- hn · 274 points · 32 comments
- hn · 292 points · 33 comments
- hn · 302 points · 33 comments
- hn · 319 points · 34 comments
- hn · 339 points · 35 comments
- hn · 355 points · 37 comments
- hn · 368 points · 38 comments
- hn · 379 points · 38 comments
- hn · 389 points · 41 comments
- hn · 403 points · 46 comments
- hn · 414 points · 52 comments
- hn · 429 points · 54 comments
- hn · 442 points · 59 comments
- hn · 447 points · 59 comments
- hn · 457 points · 60 comments
- hn · 464 points · 61 comments
- hn · 468 points · 62 comments
- hn · 472 points · 63 comments
- hn · 475 points · 65 comments
- hn · 477 points · 66 comments
- hn · 482 points · 67 comments
- hn · 487 points · 68 comments
- hn · 497 points · 69 comments
- hn · 503 points · 69 comments
- hn · 510 points · 69 comments
- hn · 513 points · 70 comments
- hn · 516 points · 74 comments
- hn · 519 points · 77 comments
- hn · 529 points · 77 comments
- hn · 533 points · 77 comments
- hn · 535 points · 77 comments
- hn · 540 points · 79 comments
- hn · 543 points · 79 comments
- hn · 570 points · 92 comments