ECE 528 Project Report
presented by (from left to right)
Mel Torrie, David Jaussi, Kerry Nelson, Bart Michaelson
A typical trumpet is 140 cm in length with a main tube bore (diameter) of 1.1 cm. The cylindrical portion is 53 cm in length and the bell diameter is 11 cm. Three piston-like valves are inserted into the cylindrical section that can change the effective length of the instrument. These valves are used singly or in combination to produce any of the different notes. Some pitch control can also be achieved by using different mouth pieces (cup, Ved, etc.). Most trumpets are keyed in B-flat.
Since Dave is the resident trumpet player extraordinaire, he went into the echo chamber (seen to the left)
and played a song 
and three tones for wave analysis. We used the echo chamber
because it minimizes the noise when recording. The baffled walls absorb the sound waves eliminating reflections
and thus delivering a pure sound sample.
These sounds were picked up by a microphone, sent through a mixer, and recorded by the Sony DAT seen to the right.
We used the DAT because it samples in digital and has much better frequency response than analog tape decks.
The DAT sampling frequency was 48 kHz.
We then transfered the trumpet tones to the Pentium computer seen to the left by hooking the DAT's analog outputs to the
analog input on the Pentium's sound card. We would have used a direct digital transfer, but the software drivers
for the digital sound card were not operating. The signals were then sampled using the Wave SE software package.
As seen on the left we sampled three separate tones to examine differences in waveforms versus frequency.
The low tone was a middle C (Bb on the piano), the mid tone was a C one octave higher, and the high tone was a
high G. We chose these particular notes because they all are played with open valves. These waves seen are windows
of the sustain portion of the envelope. The lower the tone the longer the period and the more ripples observed.
This is the time envelope of our middle tone. In our study of synthesis of trumpet sounds, we found that the attack
is the hardest part of the waveform to reproduce.
We reproduced our sounds using two different keyboards seen on the left: the Roland JX8P on the top and the Yamaha DX7 on
the bottom. We focused our attempts on the mid tone. Click for a sound sample.
You can also click here for the frequency analysis "waterfall" plot of this sound.
After that, we changed some parameters by trial and error, comparing our synthesized tone to a real tone played by Dave.
Some of the parameters that made a big difference in the tone were the cutoff frequency of the low-pass filter, the envelope
generators, and the cross-modulation between the two oscillators. The final patch is shown to the right.
The waveform to the left is the time envelope generated by the Roland. As a group we felt it sounded quite realistic.
The attack does not closely resemble the real thing, but sounds better than it looks. To get an idea of what we heard
click for the reproduced sound recorded using the DAT.
You can also click here for the frequency analysis "waterfall" plot of this sound.
The wave in the sustain portion closely resembled the square wave we started with. We were quite surprised
at how close the period came to the sampled mid tone, though it didn't have as many ripples.
The Yamaha DX7 is an FM synthesizer. We used a patch editing program on the Macintosh computer seen to the right to create
an FM patch. We patterned our patch after the actual DX7 trumpet patch which utilized one carrier with five modulators.
This gave a fair trumpet sound, but not as good as our analog patch. To compare for yourself click .
You can also click here for the frequency analysis "waterfall" plot of this sound.
We did not know exactly what parameters to alter to give the desired effects. The final patch is shown below.
The time waveform produced by the Yamaha is shown on the left. We obviously did a much better job of reproducing the attack with the DX7 than with the JX8P.
The period obtained with the DX7 was also very close to that of out original mid tone. While DX7's waveform isn't quite as square at the beginning as the original sample, it does have more of a ripple than was obtained with the analog synthesizer.
We would use three separate sampled tones to synthesize our trumpet: a low tone, a middle tone, and a high tone. Since the trumpet's range is about 2.5 octaves, this would give adequate low, mid, and high responses. These sampled tones would be stored on an external memory chip as 8-bit words. An external microprocessor would then control the reading sequence of these samples for the M114S.
For your listening pleasure, we recorded a wave-table trumpet tone ,
though it was not generated by the M114S.
From using the Roland JX8P, we learned a lot about how oscillators, filters, envelope generators, and cross-modulation affect the sound of an analog patch. In using the Yamaha DX7, we learned how important it is to have a framework to build upon. We learned the value of trial-and-error, and how nice it was to be able to use a computer to edit a patch.
Wave-table based devices are the new norm for digital sound reproduction. We all regret not having the time or the instruction required to recreate sounds using wave-table synthesis.
Overall, we enjoyed the class and this project. To sum things up, here are all of the samples we worked with.
Song
Real Tone
Roland JX8P Tone
Yamaha DX7 Tone
Wave Table Tone
Created for ECE 528 by David Jaussi, Bart Michaelson, Kerry Nelson, and Mel Torrie.
Last modified on 12-14-95.
