search

espressif / esp-tflite-micro

TensorFlow Lite Micro for Espressif Chipsets
Apache License 2.0
394 stars 84 forks source link

Use as Wakeword engine (TFMIC-35) #92

Open garudaonekh opened 3 months ago

garudaonekh commented 3 months ago

Hi, Is it feasible to use this as wakeword engine for esp32c3 instead of Wakenet of esp32-SR. I am concerned about resources consumption and size and other compatible issue. AFAIK, espressif Wakenet doesn't support self training of wakeword as of now.

Thanks;

vikramdattu commented 3 months ago

Hi @garudaonekh what's your expected model size (in parameters) and the memory you can spare. Though wake-word models are lighter on resources, it is worthwhile knowing your requirements, given, esp32c3 does not come with AI instructions and no SPI-RAM.

garudaonekh commented 3 months ago

Hi @garudaonekh what's your expected model size (in parameters) and the memory you can spare. Though wake-word models are lighter on resources, it is worthwhile knowing your requirements, given, esp32c3 does not come with AI instructions and no SPI-RAM.

Sorry, I mean ESP32S3. My end goal is to train my own wakeword in my language. and to get this git to work as wakeword engine

thirstyone commented 1 month ago

same question, but this time esp32c3 with INMP441...