Adding voice interactivity to your device/project/product is something that many people would love to do. In 2016 Seeed, with the help of our Kickstarter backers, pushed out the ReSpeaker Core and the ReSpeaker Mic Array (discontinued). Over time we began to realize two things: first, that we were very interested in continuing to pursue this area of voice interaction, and second that we needed to have solutions that we could customize for customers who wanted them to scale. We released the 2-mic array and 4-mic array HATS for Raspberry Pi, enabling makers on a budget to get into the voice interface world and learn what is possible. However, for the second goal we began to look into new solutions for developers. This is resulting in the the ReSpeaker Mic Array v2.0, and ReSpeaker Core v2.0.
Introducing the ReSpeaker Mic Array v2.0
The ReSpeaker Mic Array v2.0 is a hardware-based solution for creating or adding voice-interface to your projects. The board is a far-field microphone array device capable of detecting voices up to 5m away even with the presence of background noise. The onboard hardware-enabled algorithms are powered by the XMOS XVF-3000. It is compatible with linux, macOS, and Windows.
What has changed?
There are three major changes, all stemming from the XMOS XVF3000:
The first major change is in the quality of the algorithms. The ReSpeaker Mic Array v2.0 continues to have the onboard hardware enabled audio processing algorithms that its predecessor had – that hasn’t changed. What has changed is the quality of those algorithms. Specifically the AEC (Acoustic Echo Cancellation) algorithm received quite an upgrade. This is now a full-duplex solution, meaning that the device can continue to play audio while also listening to its environment. Previously the audio had to be paused in order to properly record. It also enables barge-in support. Barge-in supports new voice/sounds when they enter the environment, providing them a chance to be heard, even when noticeably quieter than other sounds.
Second, we reduced the microphones from 7 to 4. The reduction in microphones has allowed us to reduce the cost of the board. However, with the updated algorithms stated above, the final voice processing quality has not dropped, but instead improved. Below is a demo of the AEC in action. You can hear the device playing music as our engineer speaks:
Here is the audio file the device was playing:
Here is the end result post-processing:
Last, this board is able to be customized. The new chipset allows for both a circular array and linear array to be designed, as well as the ability to change the spacing of the microphones. This is great for fitting the solution into your products. We are looking forward to including this chipset into our OPL and also offering our own customization and manufacturing services for customers.
We hope you enjoy the new version of the ReSpeaker Mic Array (shop now). For more technical information check out our wiki. If you have questions or comments please post them on our forum. For bulk sales or distribution requests contact firstname.lastname@example.org.