Hi,
Trying to implement active voice suppression using rnn on OM13098.Used the demo application dmic-12c-dma as a starting point.But what I have understood is any delay more than 5 milli second while transferring data to codec causes the output signal to be noisy.Not an audio expert so not sure if I am missing some basic stuff here.Please do point me in the right direction
Hi xiangjun,
Adding delay in the callback also has the same effect.So for now I'm avoiding audio output.Is there any source or project that is present where keyword detection is done on this board.
Hi, Vishal,
Regarding the speech recognition based on LPC family, I have asked the AE team, we have been developing the application, but we have nothing we can share with customer till now.
I am sorry for not helping you
BR
XiangJun Rong
Hi, Vishai,
Regarding your question that "any delay more than 5 milli second while transferring data to codec causes the output signal to be noisy", as you know that the DMIC and I2S are data stream module, if you set up delay, the stream maybe interrupted, or distorted. There is two calledback function, you can handle the stream data in the callback function without setting up break point or delay while the LPC is running.
void dmic_Callback(DMIC_Type *base, dmic_dma_handle_t *handle, status_t status, void *userData)
{
if (s_emptyBlock)
{
s_emptyBlock--;
}
}
void i2s_Callback(I2S_Type *base, i2s_dma_handle_t *handle, status_t completionStatus, void *userData)
{
if (s_emptyBlock < BUFFER_NUM)
{
s_emptyBlock++;
}
}
Hope it can help you
BR
XiangJun Rong