Description

Group leader was Gary Sikorski. Main architect was Scott Foster, helped by Steve Saunders. Sam Nicolino was hardware designer and translated the achitecture into an MSI simulation and then a chip. John Palevich did some of the software, Tom Zimmerman did even more. Some original software still exists on some demostration and development disks that came with one 65XEM prototype. The disks are available at John Hardie's site.

With the firing of the Advanced Technologies Group at Atari, the new engineering staff now under the leadership of the Tramiels attempted to wedge this chip into an Atari 8bit, hence the Atari 65XEM. Shown at the 1985 CES, the 65XEM was touted as having remarkable capabilities for sound/voice synthesis. However without the original designers, getting the chip to function properly became an expensive burden which the Tramiels could not afford. This information contradicts the one from Sam Nicolino (hw designer) who said: 'I worked with them as needed.', while finishing the chip design. Sam left Atari after he gave Amy-B step design to Leonard (I assume Tramiel, ed.).

The Amy chip and technology were sold to a Millwaukee based audio design house called Sight & Sound. The company was able to not only decypher the workings of the Amy, but created a newer and more powerful version which they intended to market. However Atari suddenly reappeared on the scene and initiated a law-suit that apparently was so frightening that many of the former S&S employees that have been interviewed for the research on the Amy chip would only talk under anonymity. The Amy chip, its technology and its enhanced replacement would not only never be used by S&S, but Atari never utilized the technology either. The designs, equipment and chips lanquished in a warehouse until 1998 when the company was liquidated and everything was sold off or destroyed.

Scott Foster speaks:

Although 64 oscillators were available, only 8 frequency ramps were provided as they were harder than amplitude ramps. (This turned out to be one of my errors, I believe.) So the idea was that a 'voice' had a single frequency generator and then some arbitrary number of oscillators attached as harmonics, each with its own amplitude ramp generator but slaved to a multiple of the fundamental frequency. As a result, Amy was awesome at single voice with all 64 harmonics attached to it. This is basically high quality additive synthesis and almost has to sound good as long as the amplitudes are right. Further, the amplitudes are easy to get from an actual recording via sort-time FFT analysis. When a piece of audio is analyzed for this purpose you have to be a little careful in the case of vibrato or glissandi and this was the focus of my work at CCRMA, following up on the landmark work of Andy Moorer.

To keep the ramp generators compact, amplitude and frequency ramps were done in log space, that is, dB and semitones. I can't remember the resolution we used, Sam would probably remember. This required a log to linear conversion, we used a table lookup as I remember. Another advantage is the compact communication over the 8 bit bus to the microcontroller (we were expecting an 8051-like part).

The controller transmitted both a target (amplitude or frequency) and a slope. Amy then ramps the appropriate parameter according to the slope commanded until it hits the target.

Clearly this sum-of-sinusoids or additive synthesis approach is flexible and efficient for narrowband voices such as most musical instruments and singing voice. However, for wideband instruments such as cymbals or unvoiced speech Amy had the ability to use some of the 64 oscillators in a wideband mode. In this mode random noise was generated using maximal-length-sequence generation which was filtered to create variable bandwidth noise. The noise was then used to modulate the oscillator, shifting the center frequency of the noise to that of the oscillator. It worked ok and was a fairly unique feature, one of the main claims of the patent application we filed. (I believe this patent was allowed to die by S&S.)

About some samples of Amy recordings:
No, I don't think there are any recordings left from our Amy simulations (we had a PDP 11-45 with a way cool for its time 16-bit, 50 kHz A/D-D/A system) which is too bad 'cause I processed Michael Jackson singing Hello Dolly when he visited our lab one day.

Was the Amy needed by Atari?
Amy did not result from mainstream Atari product requirements, unfortunately. Like almost everything at Atari Reasearch, Amy was somewhat ahead of its time and represented what we though ought to happen rather than what made sense for Atari's current product line. Another huge blunder that I regret to this day.

John Foust speaks:

Our company was developing a digital additive synthesizer MIDI rack-mount based on the Atari Amy chip we'd bought from the Tramiels.

The Amy chip produced a digital audio stream by digitally summing sine waves from its ROM look-up table. It had 64 digital oscillators, essentially. Each was programmed with slope and destination: get to this amplitude in this amount of time. As such, it was an extremely low bit-rate device, in sharp contrast to other dedicated audio chip sets that needed either large amounts of RAM for samples or more frequent attention from a processor or coprocessor.

I'm still struck by the similarities of this method and today's MP3 encoders, which have at their core of tricks the same sort of frequency analysis on short segments of sound.

In one experiment I remember being able to recreate telephone-quality voice audio at a bit rate that could be pushed through a 2400 baud modem. We could analyze and store a sound such as a sampled piano key strike in just a few K, and it was pitch independent. (Of course, for a very faithful piano sound, we'd sample a key in each octave or so.)

Atari internal mailing list from 1984

From: KIM::VICKERS
Subj: Sound Chips

3) Amy

	Price: around $8.

	Performance!: Additive synthesis, 64 harmonics, which can be divided up 
		in various ways among 8 channels.  Built-in amplitude and 
		frequency envelopes.  Choice of sine or noise for the 
		harmonics.  16-bit output, you can choose how many bits you 
		want to use for your DAC.  Easy to analyze a sound and derive
		the additive synthesis parameters to resynthesize it.  Can
		do high quality speech and singing, though not efficiently.  
		A slight memory and processor bandwidth hog, but very powerful
		and flexible.

	Availability: Estimated 1st silicon in June, production quantities in
		December.  As soon as we have 1st silicon, I will begin
		recommending for people to design Amy into their games.

	Support: The sound group at Corporate R&D is designing a development
		system for Amy.  An Amy simulator is available now and will
		be moving over here before long.  Some sound analysis and
		editing software has been written.  The Amy development system
		will be much more powerful than that for the Yamaha, although
		currently it is aimed more at creating music voices than at
		sound effects (the same is true of the Yamaha support tools.)
		Amy will also need a driver to be written for RPM. 

From:	KIM::BRAD            7-MAR-1984 10:39  
To:	@SYS$MAIL:ENGINEER
Subj:	AMY 1 sound processor pricing.


The following is a memo from George Wang (ASG) and Sam Nicolino (R&D)
regarding AMY 1 chip pricing.

Excellent ASG chip plan and layout design activities have resulted in a
significant reduction in AMY 1 chip size. This results in an average sales
price (ASP) drop of $3.00 per part! Therefore, the current ASP for the
AMY 1 chip is revised to $8.82.

If you have any question about the specifications, please contact Sam
Nicolino at 745-2734 or George Wang at 745-2119.

Antic vol.11 no.3

Priced at approximately $30-50 above the 65XE is the 65XEM, which is a standard model XE with one very important addition--the AMY VLSI chip. Calling it an "advanced music synthesizer on a chip," Kerr tagged the advent of AMY as a "breakthrough," and said that the custom processor is capable of "symphonic quality sound." It features up to eight fully independent multi-timbre voices comprised of 64 separate and software-configurable oscillators, which can easily change sonic characteristic on the fly. AMY handles 10.75 octaves, from 4.8 Hz to 7.8 KHz which encompases the dynamic range of an 88-key piano.

In contrast to more conventional sound synthesis devices like Commodore's SID chip, AMY is based on digital sampling. This approach "recreates" a waveform rather than attempting to imitate it. AMY has a 30KHz digital sampling rate and a 60 dB dynamic range, and runs off the internal clock of the XEM at 7MHz.

Those who have heard AMY confirm that it can realistically portray a wide range of musical sounds. Listeners report, for example, that AMY simulates the highly distinctive "bow attack" of the cello sound with almost discomfiting accuracy. Kerr sees the 65XEM as both a music learning tool and a performance instrument.

Compute! 1985/22 April 1985

The 65XEM, essentially a 65XE with an eight-voice AMY sound chip that includes 64 oscillators. Reportedly, the AMY chip can be programmed to simulate any musical instrument. This machine will be offered for under $200.

Catalogue numbers

unknown

Pinout

Sources