Vox Probe 10

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Andrew Davie

**AtariVox / SpeakJet emulation: VoxProbe 10, please record**

Here is the tenth probe ROM. Rather than covering many loose ends a little, it asks one question thoroughly: how the SpeakJet's envelope oscillator works, and how the chip's own speech engine uses it. It runs about 3 min 20 s. As before, I'd like it from both units if possible: the original AtariVox and the AtariVox+.

**Recording.** The same setup as VoxProbe 7 to 9: 48 kHz, 16-bit, no input processing (no noise reduction, no AGC). The probe starts with 2 s of silence and then a 1000 Hz tone; set the level so that tone sits at about -10 dBFS and nothing clips. Start recording, press FIRE (or RESET) to start the sequence, and leave it running until the closing tone and silence. Each section is announced by a pair of touch-tone digits, so the recording can be matched back to the ROM.

**Why this question.** The envelope oscillator is the sixth oscillator in the SpeakJet's synthesizer. It switches the other oscillators on and off at the voice's pitch, so every vowel, nasal and voiced consonant goes through it, and so do the buzzing effects (the robots, C5, the A and C series). The emulation's model of it is the one thing all the remaining errors point back to. For example, the robot effects turn out to carry a pulse that follows Pitch at a fixed offset, which the emulation doesn't have. So the probe drives the oscillator directly through the serial protocol, then plays the speech engine's own sounds that use it, so the two can be matched.

**What it tests**

1  reference          2 s of silence, the 1000 Hz tone
2  the four shapes    saw, sine, triangle and square, each at 20, 50, 100, 200 and 400 Hz, on one tone
3  the pulse itself  each shape over a silent oscillator (does the pulse sound on its own?), and the saw
                      over tones from 250 to 3500 Hz
4  levels inside it  the oscillator's volume 2 to 16, and the master volume, under the pulse
5  its rate          the envelope's rate on a fine grid round 88, 100, 120, 150, 160, 200 and 250 Hz
                      (the chip rounds its pitch periods in two different ways, depending on the sound)
6  combinations      two and three oscillators under the pulse; the noise oscillators under it
7  the engine's use  R1, R0 and C5 at Pitch 0 to 250 in nine steps, R2 and A0 at a few, slowed, each twice
8  the voice's use    AW at Pitch 0 to 250, slowed, each twice
9  reference again    the 1000 Hz tone, 2 s of silence

SpiceWare

#1
Attachment removed, redoing capture.

The recording time for original AtariVox was significantly longer than the AtariVox+ so I probably picked the wrong probe # for one of them.

Maybe for future probes have it show the probe # as part of the "ATARIVOX TEST" message.

SpiceWare

Did it again so I hacked the ROM via Stella to confirm I'm not picking the wrong one.

SpiceWare

Probe 10 results from AtariVox+ Vectrex Enhanced

SpiceWare

#4
Probe 10 results from original AtariVox

There's definitely something going on, I confirmed the "10" was onscreen and the original AtariVox' output was still an additional 1 1/2 minutes.

4:14 AtariVox+ Enhanced
5:48 original AtariVox

Note: had to zip the recording, was too big to attach.

SpiceWare

Probe 10 results from AtariVox+ Vectrex Enhanced being driven by Stella.



I found my AtariVox USB Interface and issued this command in the terminal:

ls /dev/cu*
Then plugged in the interface and issued the command again. 2 devices showed up:

  • /dev/cu.usbserial-4
  • /dev/cu.usbserial-AVR2N0HW

Stella worked when configured with /dev/cu.usbserial-4 but not the other one.

I'm curious on the capture quality.  The capture device is plugged into a monitor while the AtariVox Interface is plugged into a HUB, so there might be some issues that would be better if they were both plugged into the same place.

Also note it's possible flow control didn't work as this recordings a little shorter than the one take on my 2600.

Andrew Davie

#6
Quote from: SpiceWare on 03 Nov 2026, 11:30 AMProbe 10 results from AtariVox+ Vectrex Enhanced being driven by Stella.

TY.  Initial analysis: "The Stella chain is much cleaner, with no hum and a lower background, so it suits the fine pulse-shape work."

VoxProbe 10's first answers (all three takes agree to 0.1 Hz):

The period routine has twelve exceptions. The envelope oscillator's own register shows the same one-tick-long exceptions VoxProbe 2 found in Pitch, so it's one routine. The default pitch is the manual's 88, and that mystery is closed.
The saw envelope sounds on its own over a silent oscillator, a direct source for the voice's low end. Sine, triangle and square don't.
Saw versus the other shapes: the saw runs floor + 1 at exact divisors, while sine, triangle and square run exact. That looks like the real basis of the "speech versus effects" period rules.
Recording and running notes:
The Stella chain has no hum and a 9 dB lower background, ideal for fine work.
On units A and Stella, the chip reset partway through the slowed section 7 (Speed back to 114). Unit B, as on VoxProbe 1, never resets, so its take is the complete slowed set.

Andrew Davie

The core question has its answer.

**What VoxProbe 10 established** (facts doc §18):
- **The envelope oscillator's period routine** runs floor(8000/rate) + 1 for the saw, with twelve rates one tick long. It's the same routine the speech Pitch uses, which closes the default-pitch question: 88, as the manual says. The fix is shipped and committed (C `3533fbb`, Go `dadf0db`): overall P 0.939 → 0.935, E 4.997 → 4.988.
- **Its shapes:** the saw is a step plus a one-signed linear fall over about 56 ticks. Sine, triangle and square are exact at whole divisors: a raised cosine, a rising ramp, and half on.
- **The engine's use in effects:** each effect's pulse runs at Pitch + offset, down to Pitch 0, rounded through the period. Its energy falls on the pulse harmonics nearest the pulsed tone, exactly the structure now shipped for R0-R7.
- **The leftover residual** in the voice and effect pulses is a fixed ±2×-rate amplitude detail at about −27 dB, not a missing mechanism.

**Recording notes:**
- Units A and Stella reset mid-way through the slowed section, while unit B didn't (as on VoxProbe 1), so unit B's take is the complete slowed set.
- The ROM sends about 60 bytes a second, one per frame, which had shifted my first placement.
- The Stella chain has no hum, but its clock reads ×1.0142, so its recording device runs at a slightly different sample rate.

**Corrected along the way:** "the saw sounds on its own" was a misplacement artefact and is retracted.