Cerebro Voice

The silent speech recognition device

Public Chat
Similar projects worth following
Cerebro Voice is a device that collects EMG signals to produce silent speech recognition.What’s silent speech recognition (a.k.a. subvocal recognition)?

The ability to recognize words without the person needing to pronounce said words out loud. This is done by collecting electrical signals via sEMG (surface EMG) directly from vocal chords and other muscles involved in voice articulation.

First results in this field was shown by NASA Ames Laboratory back in 2004 and the name they gave this technology is "synthetic telepathy" ;)

Addressing Problematic Areas:

Google Assistant, Alexa, Siri - we are currently witnessing the rise of personal AI assistants. But there are problematic issues in using solely voice recognition.

  1. Speech Recognition Accuracy - It's hard to isolate a person's voice in noisy environments.
  2. Privacy - Anything you ask Siri, everyone else hears too.
  3. Personal Identification Authentication - with AI helping with things that involve monetary transactions (purchases and payments, booking flights), we need to trust that AI tech securely identifies individuals 

Collecting non-audio signals as another stream of data when performing voice recognition tasks is a way to help mitigate these issues. 


OpenBCI casing

Standard Tesselated Geometry - 4.57 kB - 10/05/2018 at 01:29



OpenBCI casing

Standard Tesselated Geometry - 3.79 kB - 10/05/2018 at 01:29



ESP8266 casing

Standard Tesselated Geometry - 11.41 kB - 08/27/2018 at 01:30



ESP8266 casing

Standard Tesselated Geometry - 55.36 kB - 08/27/2018 at 01:30



Sensor board casing

Standard Tesselated Geometry - 9.07 kB - 08/27/2018 at 01:30


View all 8 files

  • Op Amp Testing

    Annie Ho10/22/2018 at 13:37 0 comments

  • Checking Out Flexible PCB

    Annie Ho10/22/2018 at 13:35 0 comments

    Checking out flexible pcb options when we're ready to custom design our own pcb for a wearable form factor

  • Testing Op Amps

    Annie Ho10/22/2018 at 13:30 0 comments

  • No Function Generator, No Problem

    Annie Ho10/22/2018 at 13:25 0 comments

    Generating a sine wave from Youtube in place of a wave generator

  • Developing Active Sensors

    Annie Ho10/22/2018 at 13:24 0 comments

    Initial testing with the Olimex active sensor

  • From 2-channels to 4-channels

    Annie Ho10/08/2018 at 19:31 0 comments

    We've been working on the mobile prototype and have moved from MyoWare to OpenBCI as our biosensing unit to increase the number of channels from a 2-channel board to a 4-channel board. The more sensors and channels we can get data from will increase the predictive power of our model. 

    A basic case for this unit was designed and uploaded as Files.

  • Adding casings for the sensor board

    Annie Ho08/27/2018 at 01:26 0 comments

    With a decent sensor board in place giving us some predictive power, Taylor designed new casings for the portable prototype. 

  • Successful EMG sensor board trial

    Annie Ho08/26/2018 at 03:58 0 comments

    Our final choice was MyoWare for the sensor board. We were able to get 2 channels for EMG signal + ground that gave us some predictive power. 

    Please check out our Github repo. Wesley uploaded some code for the Arduino side and Python code for the receiving side being placed on the workstation.

  • Improving the portable prototype with 'EMG CLICK'

    Annie Ho08/26/2018 at 03:49 0 comments

    Our next step was to try 'EMG CLICK' as an EMG sensor board. Unfortunately we did not have any luck with this. Perhaps we did not put enough time into exploring this. 

  • 3D modeling casing for portable prototype

    Taylor Yang08/26/2018 at 03:23 0 comments

    After verifying the proof of concept with stationary data acquisition jig, we are ready to move on to prototyping a portable solution. We started 3d printing enclosures for the WiFi chip ESP8266 and microphone amplifier KY038.

    I have 3D modeled several versions and 3D printed to try out.

    Didn't want to add too much complexity to the prototype so just used a simple top and bottom piece which clips together, while the wires will be run from the back.

    Result: It gave mediocre results for sound acquisition and results were not great for converting the microphone amplifier KY038 into an EMG amplifier.

View all 10 project logs

  • 1
    Place electrodes where voice muscles are

    Concentrate around the throat area and Adam's apple but do not put pressure directly on Adam's apple

  • 2
    Connect electrodes to an pre-amplifier data acquisition rig
  • 3
    Use the pre-amp rig to collect the EMG signal data

View all 7 instructions

Enjoy this project?



wolfgangouille wrote 03/11/2020 at 22:04 point

Is this actually working? I am building a sEMG wristband, it is so sensitive that i can control a device without actually moving. But I always thought it is not possible to record the vocal chords controlling muscles without invasive electrodes.

  Are you sure? yes | no

Carl Bugeja wrote 09/16/2018 at 21:07 point

Had a similar idea once but gave up on trying to filter out the human thoughts/inner-voice.. you should try to combine it with a bone conduction speaker ;)

  Are you sure? yes | no

Curt White wrote 09/15/2018 at 00:06 point

This is very interesting. I read your GitHub repo README. I got as far as storing and labeling data. What kind of processing are you doing for voice recognition ie the relationship between data? How to you measure the quality of the data?

  Are you sure? yes | no

Shervin Emami wrote 09/05/2018 at 22:34 point

That's quite an impressive DIY project! Silent voice recognition has a huge commercial potential, but only if it's done reliably and fairly conveniently. I know there has been some research into silent voice recognition but this is the first time I've seen info for a DIY version :-) Have you made comparisons of how reliable or accurate your system is compared to traditional audio based speech recognition?

  Are you sure? yes | no

Similar Projects

Does this project spark your interest?

Become a member to follow this project and never miss any updates