"Dan Ash" <dan_ash@sbcglobal.net> wrote in message
news:1138027041.979088.172920@g49g2000cwa.googlegroups.com...
> For note detection, a 16-bit processor is adequate. For multi-effect
> processing, 24-bit, 32-bit or floating point is necessary to battle the
> noise floor. People like the output quiet when not playing.
>
> You want to purchase an evaluation board to do experiments and the TI
> C5xxx, Analog Devices Blackfin, and Motorola 563xx/568xx processors are
> supported by stand-alone evaluation boards. For floating point, the TI
> C6xxx and Analog Devices SHARC processors are supported by the
> evaluation boards. All evaluation boards are between $300-$600 US and
> come with Windows development software for C/asm programming and
> debugger controlled via USB.
>
> I am not sure if you want to stream samples or just detected notes to
> the PC. Some of the boards have on-board RS-232 support. TI and
> Analog Devices daughtercards with USB peripherals for streaming sample
> data.
>
> Regarding algorithms, Wavelet, FFT, prefiltering (high-pass to get rid
> of harmonics on the high strings) and counting zero-crosssings are all
> workable techniques. Expect harmonics.
>
> Good luck,
>
> Dan Ash
>
> Forget about FFT for single note detection, especially if you want to
> detect transients reliably.
> And I have no idea who suggested to use Wavelets for this purpose: I
> haven't seen any useful
> implementation of Wavelets for pitch detection (Correct me if I am
> wrong...)
Can you suggest another more effective way for us to implement pitch
detection? I am particularly concerned with being able to handle
transients.
>
> On the subject of FFT, somebody (Georg von Bekesy ?) once said:
> "Dead cats and Fourier transforms have harmed hearing science more than
> anything else".
>
I have never heard of that quote but that sure is interesting! :)
Reply by fizteh89●January 24, 20062006-01-24
>Our university team is restricted to a budget of 375 USD, is there
>anything in the sub 200 USD range that can satisfy my requirements?
EBay is your friend...
>I think we will stick with FFT because we don't have the expertise to
>come up with algorithms for Wavelet transforms. It also seems that the
>wavelet transform is more computationally intensive.
Forget about FFT for single note detection, especially if you want to
detect transients reliably.
And I have no idea who suggested to use Wavelets for this purpose: I
haven't seen any useful
implementation of Wavelets for pitch detection (Correct me if I am
wrong...)
On the subject of FFT, somebody (Georg von Bekesy ?) once said:
"Dead cats and Fourier transforms have harmed hearing science more than
anything else".
Reply by Brian●January 23, 20062006-01-23
Thanks a lot for your reply that was very helpful.
> For note detection, a 16-bit processor is adequate. For multi-effect
> processing, 24-bit, 32-bit or floating point is necessary to battle the
> noise floor. People like the output quiet when not playing.
The requirements for my project have changed slightly, the DSP will not
be outputting any audio, it will simply detect notes and based on the
note, it will output control signals to a decoder logic circuit that
will control some LEDs.
> You want to purchase an evaluation board to do experiments and the TI
> C5xxx, Analog Devices Blackfin, and Motorola 563xx/568xx processors are
> supported by stand-alone evaluation boards. For floating point, the TI
> C6xxx and Analog Devices SHARC processors are supported by the
> evaluation boards. All evaluation boards are between $300-$600 US and
> come with Windows development software for C/asm programming and
> debugger controlled via USB.
Our university team is restricted to a budget of 375 USD, is there
anything in the sub 200 USD range that can satisfy my requirements? We
have found a chip from Silicon Labs, the C8051F120 that fits our price
range but seems a lot less powerful than the chips you suggested. Can
we get away with using the C8051F120?
> Regarding algorithms, Wavelet, FFT, prefiltering (high-pass to get rid
> of harmonics on the high strings) and counting zero-crosssings are all
> workable techniques. Expect harmonics.
I think we will stick with FFT because we don't have the expertise to
come up with algorithms for Wavelet transforms. It also seems that the
wavelet transform is more computationally intensive.
> Good luck,
Thanks again for your feedback.
Reply by Dan Ash●January 23, 20062006-01-23
For note detection, a 16-bit processor is adequate. For multi-effect
processing, 24-bit, 32-bit or floating point is necessary to battle the
noise floor. People like the output quiet when not playing.
You want to purchase an evaluation board to do experiments and the TI
C5xxx, Analog Devices Blackfin, and Motorola 563xx/568xx processors are
supported by stand-alone evaluation boards. For floating point, the TI
C6xxx and Analog Devices SHARC processors are supported by the
evaluation boards. All evaluation boards are between $300-$600 US and
come with Windows development software for C/asm programming and
debugger controlled via USB.
I am not sure if you want to stream samples or just detected notes to
the PC. Some of the boards have on-board RS-232 support. TI and
Analog Devices daughtercards with USB peripherals for streaming sample
data.
Regarding algorithms, Wavelet, FFT, prefiltering (high-pass to get rid
of harmonics on the high strings) and counting zero-crosssings are all
workable techniques. Expect harmonics.
Good luck,
Dan Ash
Reply by Brian●January 22, 20062006-01-22
Hi All,
I am currently trying to build a tool that is capable of providing
information about the real time characteristics of the output of a
electric guitar. Bascially I want to determine what note (or
combination of notes) is being played in real time, based on this info I
will give some sort of feedback to the user. It has been suggested to
me that I should use a wavelet transform or FFT to do this, but since my
knowledge of DSP is limited to a senior level course in introductory
DSP, I have little clue on how to implement this in real time.
I am planning to do this using the Motorola 56F800 series chips or the
silicon labs C8051F12X series chips.
Can anyone provide me with any guidance or input? At this point I could
use whatever help I can get.
Thanks.