is there any way of accessing the sounds that are sent to speech sdk server

Faris Lemes 50

hi, I'm trying to make some ai assistant using speech SDK, device is Linux kernel based, and I've configured Alsa loopback and Pulseaudio to utilize the echo cancellation feature which should be supported by SDK. One thing that I noticed is that sometimes echo cancellation works flawlessly, but sometimes it does not. I'm trying to figure out is there something wrong on my side. I've tried to record all channels using arecord and Alsa while I'm experimenting with sound and speaker sound is really recorded on the last channel as it is requested by Speech SDK. Which filter is used on Speech SDK side? Is it some adaptive filter or some filter which is buffering the loopback channel? Is there any way to debug this behavior? I've tried to analyze the logs that SpeechSDK generates but I could not find anything interesting there.

Thanks!

dupammi 7,955 Reputation points Microsoft Vendor

2024-06-06T10:04:32.29+00:00

Hi @Faris Lemes

Regarding the filter used by the Speech SDK, the exact details of the filter are not publicly available.

You can try using noise suppression with Microsoft Audio Stack (MAS) to see the audio does not have ambient noise. To use noise suppression with MAS, you can refer to the official Azure documentation on how to use the Microsoft Audio Stack with the Speech SDK.

I hope you understand. Thank you.
Faris Lemes 50 Reputation points

2024-06-06T11:00:04.36+00:00

Is there any known bug or issue related to the echo cancellation, because at 50% cases it works without any issues, and 50% cases it does not work at all. Is there any explanation behind this. There are no ambient noises in the sound.
dupammi 7,955 Reputation points Microsoft Vendor

2024-06-07T10:28:42.6533333+00:00

Hi @Faris Lemes

Thank you for your follow-up question.

Regarding your question about the echo cancellation feature, there are no known bugs or issues related to the feature in the Speech SDK. However, there could be various factors that affect the performance of echo cancellation, such as the quality of the microphone, the acoustics of the room, and the distance between the speakers and the microphone.

If you have already tried adjusting the settings and using noise suppression, and the issue still persists, you can try some of the other solutions mentioned in the Azure documentation on how to mitigate or resolve echo issues.

Once after trying above, if still issue persists, I request you to raise a support case through Azure portal.

I hope you understand. Thank you.

Share via

is there any way of accessing the sounds that are sent to speech sdk server