WisBlock Audio: Speech Recognition Solutions

WisBlock Audio speech recognition solutions allow you to add voice interaction to your devices. These devices could be playing music, controlling motors or lights. There are so many possible applications, it is only limited by your imagination and what you can connect into your WisBlock system. WisBlock Audio provides a number of different ways for you to add offline speech recognition when building a product with voice control. This page outlines the differences between each of the options available to WisBlock Audio developers.

WisCore with Voice Processing DSP

The RAK18080 DSPg DBM10L module is a voice processing module using the Synaptics/DSPg DBM10L chip. This DSP chip has the capability to run speech recognition software from a leading embedded speech solution provider while also performing Voice Processing (noise reduction, acoustic echo cancelation etc.) RAK provides example voice models which can be used to evaluate the solution, or even use for your project.  For larger customers RAK can create new models (different trigger word and commands) for a fee. Running the speech recognition software on the DSP means that less processing power is used on the WisCore while providing better performance than running directly on the WisCore.

Example voice models include:

• Triggers:
  • Hey RAKStar, Hey RAKCloud, Hey Helium
• Light control
  • Lights On, Lights Off, Dim Lights, Lights Red, Lights Blue, Lights Green, etc.
• Music Commands
  • Play, Stop, Next, Resume, Volume Up, Volume Down, etc.

Note: The RAK18080 is only available for purchase in a kit as the DSPs algorithms have been tuned to match the audio enclosure. It is available in the following kits:

• The WisBlock Voice Processing Kit
• WisBlock Voice Processing Speaker Kit

WisCore Licensed for Speech Recognition (Software only)

You can purchase a WisCore module that is licensed to run speech recognition software from a leading embedded speech solution provider. The software running on the WisCore is very similar to the software running on the RAK18080 and uses the same voice models (see example voice models above). For larger customers RAK can create new models (different trigger word and commands) for a fee. Running the speech recognition software on the WisCore means that no extra hardware is needed beyond a WisCore, Base and microphone.

WisCore & Edge Impulse (Software only)

Using Edge Impulse to create your own model is a low-cost solution as you aren’t purchasing a license or extra hardware. With Edge Impulse you can create any model you want and improve it until it meets your requirements (false triggers vs false rejects). Also, by using Edge Impulse you can even recognized sounds other than voices (eg a barking dog). For more information on Edge Impulse running on a WisBlock see : “Getting Started with WisBlock and Edge Impulse” Is this the correct title.. also should link to the page/documentation.

Quick Comparison Table:

Notes:

1. RAK can create models for you for a fee for large customers
2. Performance is a measure of how well voice commands are recognized (false triggers vs false rejects)
3. Although power consumption is listed it is not expected that a WisBlock speech recognition system would run off battery.