AER FDNLP corrupted output

Simon Pearson1

Prodigy 150 points

We are integrated the AER library with the following setup:

sample rate: 8kHz (all I/O)

version: 17.0.0.0

processor: c6748

We have followed the example code to create a similar simulation environment and used the example parameter set.

The AER is producing clean audio output in the send path and providing echo cancellation except when the FDNLP is enabled.

The send path output with FDNLP processing appears to consist of broken-up near end speech frames. The output audio file contains 40 samples of near end speech, followed by 40 samples of low level noise (-60dbFS, noise from the comfort noise generator?), repeating every 80 samples. The output is ok when TDNLP is used. The FDNLP issue occurs either in hands free or headset modes.

Could you provide any guidance on what might be the likely cause of this issue?

Initial checks show that the buffer allocation seem correct. Are there any particular buffer requirements or parameter setting we should check?

Kind regards,

Simon

over 9 years ago

0 Jianzhong Xu over 9 years ago

TI__Mastermind 40215 points

Hi Simon,

Sorry for my late response. Do you still experience this problem or is it resolved?

If this problem still exists, can you store the send path output into a file and send to me?

Regards,

Jianzhong

0 Simon Pearson1 over 9 years ago in reply to Jianzhong Xu

Prodigy 150 points

Hi Jianzhong

Yes we still experience this problem.

Two files of the send path output are attached, one with the FD NLP enabled and one with it disabled (TD NLP).

Both files process the same audio: 0-3s is near end speech only, 3-6s is far end speech only and 6-9s is double talk.

Thanks,

Simon

FDNLPwavfiles.zip

0 Jianzhong Xu over 9 years ago in reply to Simon Pearson1

TI__Mastermind 40215 points

Hi Simon,

This looks like some buffer placement issue. Can you try the following to narrow down the problem:

1. disable ASNR and see if signal is clean

2. disable CNG and see if signal is clean

Meanwhile, is it possible for you to send me your application code where AER is instantiated and configured?

Regards,

Jianzhong

0 Simon Pearson1 over 9 years ago in reply to Jianzhong Xu

Prodigy 150 points

For the previously attached signal, both ASNR and CNG were already disabled. Does this help?

Could you send a contact email address to send the application initialisation code to?

Thanks,

Simon

0 Jianzhong Xu over 9 years ago in reply to Simon Pearson1

TI__Mastermind 40215 points

Simon,

Please send it to xuj@ti.com.

Regards,

Jianzhong

0 Jianzhong Xu over 9 years ago in reply to Simon Pearson1

TI__Mastermind 40215 points

Hi Simon,

I've received the integration code, but didn't see anything obviously incorrect. Can you perform two more tests as described below and provide me the captures?

1. disable Tx NLP

2. enable Tx NLP and use FDNLP but disable Tx clipper (through aer_CTL0_DISABLE_TX_CLIPPER)

Please also send me the debug status returned by API function aerGetPerformance().

Thanks,

Jianzhong

0 Simon Pearson1 over 9 years ago in reply to Jianzhong Xu

Prodigy 150 points

Many thanks, will send through the captures soon.

0 Simon Pearson1 over 9 years ago in reply to Jianzhong Xu

Prodigy 150 points

Hi Jianzhong

The captures are attached with the debug status logs from aerGetPerformance().

Please let me know if you need more information.

Thanks,

Simon

FDNLPwavfiles2.zip

0 Simon Pearson1 over 9 years ago in reply to Simon Pearson1

Prodigy 150 points

Also, are there any constraints on the I/O buffer alignment and size to the function aerSendIn() for FDNLP with a sampling rate of 8kHz?

Simon

0 Simon Pearson1 over 9 years ago in reply to Simon Pearson1

Prodigy 150 points

FDNLP_16kHz.wav

This issue also occurs at a sampling rate of 16kHz when FDNLP is enabled, see Tx output attached.

For 16kHz, the Tx output contains a repeating pattern of 80 samples of signal followed by 80 samples of near silence.

With FDNLP disabled, the output looks ok.

Thanks,

Simon

0 Jianzhong Xu over 9 years ago in reply to Simon Pearson1

TI__Mastermind 40215 points

Simon,

Thanks for the information. The symptom is the same for 16kHz, i.e. half of the Tx output buffer is corrupted. A couple of questions here:

1. I saw from the debug stats that the tail length is 5. This doesn't look correct.

2. You have two AER instances. Are both running? Which one gives you this problem?

3. Did you run the test in a real time system or in a simulation environment? If you used simulation, would you mind sending me the whole project so that I may be able to debug it?

Thanks,

Jianzhong

0 Simon Pearson1 over 9 years ago in reply to Jianzhong Xu

Prodigy 150 points

One instance is configured to use aer_PHM_HES mode and one to aer_PHM_HF. Only one instance is running at a time. The same issue occurs in either instance.

I've set the tail length to the configured maximum (160) for the aer_PHM_HES instance. This does not seem related to the issue though.

This was tested in a simulation environment. Will you be able to debug the binary (.out) built with debug symbols alongside the source?

Thanks,

Simon

0 Jianzhong Xu over 9 years ago in reply to Simon Pearson1

TI__Mastermind 40215 points

Simon,

Yes, I should be able to debug the code if you send me the whole project including your integration code.

Thanks,

Jianzhong

Processors

Processors forum

AER FDNLP corrupted output