Andrea PureAudio Microphone Development Kit
The two pages you’re reading now were written on the morning of this column’s deadline. Not because I’m a terrible procrastinator – this time around, anyway – but because the original review I wrote prompted the company behind the product to completely rethink its approach to market.
Originally, you’d have been reading a scathing review of the Pure Audio Speech Development Kit (SDK) from Andrea Electronics, a long-standing name in the audio industry. Billed as the one-stop system for high-quality speech control of Internet of Things (IoT) and voice-assistant projects running on a Raspberry Pi, the SDK proved to be anything but.
During testing, the kit was found to have several issues. For starters, there was a mandatory USB sound card that was tied to the software purely for licensing purposes, having no dedicated hardware or specialist components over a £5 unit from any other manufacturer. There was also an outdated makefile that simply wouldn’t work without manual modification, and an audio filter library that can’t run echo cancellation on any Pi device below a Raspberry Pi 3.
In addition, the licence precluded the user from modifying the source code of the sample application, rendering it useless. Then there was the sample application itself, which was less a starting point for development and more a bare bones demo better suited to showing off Andrea’s audio technology at trade shows.
Far from allowing the user to develop impressive Siri or Alexa style applications, the SDK instead had only two functions: listen for a pre-set trigger phrase, ‘Hello Blue Genie’, and trigger a ten-second recording. Additional features in the form of specific commands can be added, but they’re available only in extremely limited ‘vocabulary packs’ as paid add-ons. These packs require you to state the command exactly as written, making ‘turn on the light’ a valid command but ‘turn the light on’ an invalid one, and come with no actual abilities, bar printing a message to the terminal confirming that a command has been detected.
To its credit, Andrea Electronics took the news that it had perhaps not developed the SDK to the point it had promised with aplomb – and, the day before this column was due, performed a dramatic volte-face. No longer was it the PureAudio Speech Development Kit, but the Noise-Cancelling Microphone Development Kit.
No longer was it positioned as a means for hobbyists to play with voice control, but instead a jumping-off point for experienced developers eager to experiment with Andrea’s technology. The vocabulary files, meanwhile, are still set for launch, but they’re no longer a primary focus of the product.
The real meat of the product, Doug Andrea explained in an email exchange, is the filter library. Using the audio input from the bundled stereo microphone – a design that will be immediately familiar to anyone who purchased select Asus motherboards a decade ago as an in-pack freebie – the filter library allows for relatively fine-grained control of the incoming stream.
The controls include far-field enhancement, a fancy new name for ‘microphone boost’, which amplifies quiet
signals. There’s also the flagship noise reduction feature, along with echo cancellation, which pops and crackles on any Pi device below a Pi 3.
Finally, the beamforming control attempts to ‘zoom in’ on subjects not directly centred in front of the microphone.
Testing these features, there’s no denying they work. Simulating a noisy factory setting – the sort of industrial environments into which Andrea has been selling its microphone technology for years – and toggling the noise reduction did, indeed, considerably reduce the background noise at the cost, typical of such filters, of making the recorded speech sound a little muddy and artificial.
Unfortunately, there’s a major caveat. Unlike the days when the SuperBeam was bundled with Asus motherboards, along with a Windows driver bundle for control of all these shiny filter modes, the Raspberry Pi version comes with only the basic sample application, so the only way to use the filters is to write your own application. If you’re hoping to improve the quality of your Google Hangout sessions or Audacity recordings, you’re going to be disappointed.
Even if you’re willing to write your own application, the filter library is entirely software-based, which is why the kit is only functional on the powerful Raspberry Pi 3, and eats between 30 and 40 per cent of a CPU core’s resources while running the hotphrase detection. Had the mandatory USB sound card – based on a C-Media CM6400 chip, designed for inclusion in low-cost USB headsets – included dedicated processing hardware to justify its mandatory use, the kit could have been compatible with the whole Raspberry Pi range.
The Andrea PureAudio Microphone Development Kit is available to buy from
www.andreaelectronics.com for $49.95 US (around £43 inc VAT), with vocabulary files – which you can’t make yourself, nor edit in any way to add to the very limited options on offer – priced at $9.95 US (around £8.54 inc VAT) each.
Originally the Speech Development Kit, Andrea’s Pi bundle has taken an abrupt shift in tone
The ‘SuperBeam’ microphone will look familiar to buyers of some Asus motherboards a decade or so ago
The USB sound card is a cheap C-Media CM6400 affair, but the software won’t work without it
The noise reduction filter definitely works, but you can’t use it with thirdparty applications
The ‘far field enhancement’ mode does a good job of boosting weak signals, but can’t be automatically activated