A New Approach to Pitch and Voicing Detection Through Spectrum Periodicity Measurement

Progress Report 2
Elec 499B group 11

Voice Activated Mouse

We have made much progress over the past few weeks on the “Voice Activated Mouse” project. All of the hardware schematics have been designed, built, and tested. Several of the circuits were simulated using Micro Cap 5. Frequency analyses were also obtained for many of the circuits. The frequency response plots of the analogue input circuit confirmed that it acts as a band pass filter, passing frequencies between the range of 40-5000Hz.

Over the past few weeks, we have spent much time trying to solve a noise problem that we had with a few of our circuits. We are trying to power our system through a USB port, which gives us 5V and 200mA shared with the mouse. Some components in our circuit require other voltages, so we used a capacitor charge pump voltage doubler and negative voltage power supply. We ran the power supply at 10 KHz because it operates most efficiently at that value. When switching the capacitors between series and parallel, we found that it caused current surges and voltage jumps at a rate of 10 KHz (that was amplified at a factor of 1000 by our audio amplifier circuit) giving us a very poor signal to noise ratio. In attempt to improve the signal to noise ratio, we filtered the output with an RC filter circuit using fairly large (4700uF) capacitors. When this circuit was tested, it was determined that the signal to noise ratio was even worse. After some investigation we found that the power supply was sending noise back through its power input and into the power supply for the rest of the circuits. We used another RC filter (10ohms / 100uf) to filter out this noise and obtain a much better signal to noise ratio. We discovered that the power supply produced an adequate amount of power at100 kHz, therefore, the power supply was changed to 100 KHz, which greatly reduced the amount of noise in the circuit.

When testing the band pass filter on batteries, the circuit worked fine, but when we tested it on a power supply, we got a large 60 Hz noise signal. We will have to raise our high pass filter from 40 Hz to 120 Hz.

The microphone that we had originally purchased did not work to our standards. The signal that it would detect was much too weak when compared to the noise level. As a result, we obtained an internal computer speaker to use as the input device. It was determined that this speaker had an impedance of 8 ohms. The speaker was connected to an audio transformer that had 8 ohm impedance on one side and 1.2k on the other.

A new website was created that marketed our product from a business point of view. The resolution of the pictures was decreased in order to increase the download speed. A roll-over navigation menu was programmed and tested in several browsers and at several screen resolutions. The new site did not meet our standards when viewed older Netscape browsers and when viewed using Mozilla. The Java script roll-over menus performed very poorly, therefore, we have decided to make a third website based on cascading templates, containing no flash or java script whatsoever.

Site 1: (please note that some of the drawings are different than the ones that I have attached)

Site 2: (Marketing our product from a business point of view) **please note that this site is a template that we created for testing purposes (it too lacks much of the latest documentation)

Site 3: Should be completed (designed and tested) by March 13. We will send out the new link once it is done.

Software was tested on a 16F876A pic, but since it does not have a hardware multiplier, we were unable to do any complex software routines on it. As a result, we ordered several 18 series pics. The pics have arrived. The compiler that we used for the 16F876A was not compatible with 18 series pics. We have obtained a new compiler that works with the 18 series and we are currently learning to use this new compiler.

Several voice detection routines have been investigated including; voicing detection through spectrum periodicity measurement, harmonic matching, and an instantaneous frequency method. Implementing such voice detection routines on a pic has proved to be very troublesome. We have roughly 250 clock cycles per sample to perform any necessary calculations, but when you program in C, most of the clock cycles are used to move data. We have estimated that roughly 1/4 of the clock cycles can be used foradditions and multiplications. We will be modifying one (or several) of the voice detection routines so that we can implement it on the pic. We are also contemplating a“train button”that would allow the device to target in on a particular person’s voice (based on a 1 second sample of the target person's voice).

Frequency Response of the Analogue Signal Circuit

Power Supply Circuit: