On 10/25/19 9:43 AM, Lu, Brent wrote:
On 10/25/19 4:11 AM, Brent Lu wrote:
The first DMIC capture always fail (zero sequence data from PCM port) after using DSP hotwording function (i.e. Google assistant).
Can you clarify where the DSP hotwording is done? Intel DSP or rt5514?
Turning on the MCLK with the BIAS info might force the Intel DSP to remain on, which would impact power consumption if it was supposed to remain off.
Hi Pierre,
It's done in rt5514's DSP and the interface is SPI instead of I2S for the voice wake up function.
There is a driver rt5514-spi.c which provides platform driver and DAI. User space application first uses the mixer to turn on the voice wake up function:
amixer -c0 cset name='DSP Voice Wake Up' on
Then open and read data from the PCM port which goes to the SPI platform and dai code. Finally it uses same mixer to turn off the function and return to normal codec mode. The DMIC recording (from I2S) and the voice wake on function should be mutually exclusive according to the driver design.
In the codec driver rt5514.c there is a rt5514_set_bias_level function. It's expected to turn on/off mclk here according to Realtek people's say but our ssp clock requires set rate function to be called in advance so I implement the code in machine driver.
Can you clarify if the rt5514 needs the MCLK while it's doing the hotword detection?
My point is really that this patch uses a card-level BIAS indication, and I'd like to make sure this does not interfere with the audio DSP being in D3 state.