testing ic against py_ic #266
Conversation
modules/lib_ic/api/ic_state.h
Outdated
| @@ -177,7 +177,7 @@ typedef struct { | |||
| bfp_s32_t error_bfp[IC_Y_CHANNELS]; | |||
| /** Storage for Error and error mantissas. Note IFFT is done in-place | |||
| * so the Error storage is reused for error. */ | |||
| complex_s32_t DWORD_ALIGNED Error[IC_Y_CHANNELS][IC_FD_FRAME_LENGTH]; | |||
There was a problem hiding this comment.
Is there a reason for changing this? Since this is 'Error' and not 'error', complex_s32_t seems more intuitive, although IDFT is done in place and same memory stores error so not a big deal I guess
There was a problem hiding this comment.
I did it for the tests and forgot to change it back. I agree that the complex_s32_t makes more sense with this name
| @@ -257,7 +293,8 @@ void ic_adaption_controller(ic_state_t *state, uint8_t vad){ | |||
|
|
|||
| const float_s32_t one = {1, 0}; | |||
| const float_s32_t zero = {0, 0}; | |||
| const float_s32_t delta = {1100, -40}; //1100 * 2**-40 = 0.000000001 (from stage_b.py) | |||
| //const float_s32_t delta = {1100, -40}; //1100 * 2**-40 = 0.000000001 (from stage_b.py) | |||
| const float_s32_t delta = {352, -45}; //352 * 2**-45 ~ 0.00000000001 | |||
There was a problem hiding this comment.
Is there a reason for changing this? Could you add a comment here?
There was a problem hiding this comment.
This delta is being used to prevent zero division. In the test case I saw, the denominator was 1.7e-8 while the previous delta was 1e-9. In some cases, it was significant enough to make the fast ratio more than 1 where it shouldn't be so. The test was failing with this delta. I've changed it in both C and python to the less significant one 1e-11, and the test has passed. I'll add a comment to cover that.
There was a problem hiding this comment.
That makes sense. Since we're changing an IC parameter, do ensure that the full pipeline tests keyword results don't show anything unexpected.
| @@ -213,14 +215,22 @@ def get_suppression_arr(input_audio, output_audio): | |||
| for i in np.arange(0, num_frames*frame_advance, frame_advance): | |||
| i = int(i) | |||
| in_rms = rms(input_audio[0][i:i+frame_advance]) | |||
| out_rms = rms(output_audio[0][i:i+frame_advance]) | |||
There was a problem hiding this comment.
You could do something like, out_rms = out_rms.reshape(1 , out_rms.shape[-1]) to force it into a 2D shape. Then you'll not need the if condition below.
There was a problem hiding this comment.
The output_audio can either be a single channel or have two channels. This will work fine for a single-channel array. However, I can't reshape a (2, 240) array into a (1, 240) array, python will give me an error. So I'll have to have an if ether way
There was a problem hiding this comment.
Right, that makes sense. I thought the output was either shaped (n,) or (1,n).
| @@ -273,7 +283,11 @@ def check_delay(record_property, test_case, input_audio, output_audio): | |||
| output_audio = np.asarray(output_audio, dtype=float) / np.iinfo(np.int32).max | |||
|
|
|||
| frame_in = input_audio[0][:frame_advance*2] | |||
| frame_out = output_audio[0][:frame_advance*2] | |||
No description provided.