Vox Probe 9

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

Andrew Davie

**AtariVox / SpeakJet emulation: VoxProbe 9, please record**

Here is the ninth probe ROM, replacing the one I posted and withdrew before anyone had recorded it. Three sections were added since, so this is the one to record. It runs about four minutes and, as before, I'd like it from both units if possible: the original AtariVox and the AtariVox+.

**Recording.** The same setup as VoxProbe 7 and 8: 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.

**What it tests**

1  reference          2 s of silence, the 1000 Hz tone
 2  Slow/Fast between  Slow and Fast placed before and after a sound, for seven kinds of sound: does a Slow
                       read during one sound stretch that sound, the next one, or neither? (the main timing
                       error left in Man Goes Down's long phrases)
 3  reading ahead      which commands the chip takes early, and where in a sound: pauses, Bend, Stress, Relax
                       and Volume after five different sounds
 4  effects' Speed     three effects at Speed 20 to 127 with nothing after them
 5  nasals' pulse      NGE, NE, MM repeated at Pitch 60, and at Pitch 40 with Bend 10
 6  hisses, loud       the hisses and stops at Volume 127 (the earlier probes' were too quiet to measure)
 7  robots' pulse      R5, R8, R9 at Pitch 20 and 40 (their unexplained low-frequency energy)
 8  bed against level  B0 and B1 at Volume 127 down to 16 (the hiss the effects carry: does it follow Volume?)
 9  direct tones       tones set straight through the serial protocol, with the oscillator's own volume
                       stepped, and two tones together (the harmonics that didn't follow the oscillator's
                       volume last time, and whether mixing products follow either tone)
10  Bend swept         five effects at Bend 0, 1, 2, 4, 8 and 15. Bend shifts every tone by a fixed amount;
                       what the chip does with a tone that Bend pushes to or below zero is the question. At
                       Bend 0 the recordings so far show new lines near 2-2.5 kHz for some effects and none
                       for others
11  period rounding    one effect and one vowel at Pitch 100 to 250: the chip's voicing period at Pitch 200
                       is one tick shorter than the rule fitted at lower Pitches predicts
12  near zero          a direct tone at 0, 1, 2, 4, 8, 16, 32 and 64 Hz: what a tone at next to nothing
                       sounds like (at Bend 0 and 15 a silent oscillator seems to give a 1 kHz buzz)
13  reference again    the 1000 Hz tone, 2 s of silence

Sections 10 to 12 are the new ones. They come from a day spent on the effects at Bend 0 and at high Pitch, where the emulation is furthest off and the recordings I have can't settle what the chip is doing, because every effect plays several tones at once. These sweep the settings finely on a few chosen sounds, which should make it unambiguous.

Thanks again for the recordings; everything the emulation knows about the chip has come from them.

SpiceWare

Probe 9 results from AtariVox+ Vectrex Enhanced

Same setup as probe 8, which is the same as for 7 with the added RF core. Results are almost 5 minute.

I did see it, but was on my iPad where I'm not logged in. By the time I fired up my Mac it was gone.

Andrew Davie

Quote from: SpiceWare on 02 Nov 2026, 02:41 AMProbe 9 results from AtariVox+ Vectrex Enhanced

Same setup as probe 8, which is the same as for 7 with the added RF core. Results are almost 5 minute.

I did see it, but was on my iPad where I'm not logged in. By the time I fired up my Mac it was gone.

TY. Had some processing running in the background and it had built up a bunch of questions for probe 10. I saw the opportunity to combine with 9 and since it was not marked as downloaded yet I thought I'd save an iteration.

Andrew Davie

Quote from: SpiceWare on 02 Nov 2026, 02:41 AMProbe 9 results from AtariVox+ Vectrex Enhanced

One was attached; assuming the other is coming soon but just mentioning the ambiguity of "results" in this post.

SpiceWare

Probe 9 results from original AtariVox

SpiceWare

Quote from: Andrew Davie on 02 Nov 2026, 02:47 AMambiguity of "results"

I've been using "results" each time - it's common for us to say "the test results"

It could be an American english vs British english thing like how we use "math" instead of "maths" as the shorthand for "mathematics".

SpiceWare

There's also this limit for attachments, which most of the probes would exceed if I attach both recordings to the same reply:

QuoteRestrictions: maximum total size 60 MB


Andrew Davie

TY @SpiceWare for the recordings!  Most useful.

# VoxProbe 9: findings so far (2026-11-02, both units, first pass)

Recordings: `recordings/vox_probe9.wav` (unit A, AtariVox+) and `original_atarivox_vox_probe9.wav` (unit B), converted losslessly from the .aifc. Both 48 kHz 16-bit, the right channel carrying the signal, no clipped samples, peaks -2.0 and -1.7 dBFS, the 1 kHz reference at about -6 dBFS, background about -42 dBFS (the same as probes 7 and 8).

Placement (`p9lib.py`, probe 8's method): all 13 sections found on both units; sections 1, 2, 4, 5, 7, 10, 11, 13 placed item for item; 3, 6, 8, 9 and 12 have run counts that don't match the items (merged or quiet items) and need hand placement.

## Clocks (section 1 and 13, the SCP 1000 Hz tone, 1 s, 0.05 Hz resolution)

unit A   1005.80 / 1005.89 Hz  ->  clock x1.0058, tick rate 8046.8 Hz
unit B   1001.59 / 1002.00 Hz  ->  clock x1.0018, tick rate 8014.3 Hz   (ratio B/A 0.996, as measured before)
[/clock]
`
The engine's RATE 8026 is the mean. Every tick count below uses the unit's own clock.

## The voicing period (section 11), settled apart from the default pitch

C5's buzz is flat through its last part (no drift) and both units give integer periods:

[code]
                 Pitch   100   120   150*  160   200   250    default ("88", VoxRecord and probe 4, 4 takes)
effect C5        ticks    80    67    54    50    40    32    92
floor(8000/p)+1           81    67    54    51    41    33    91
ceil(8000/p)              80    67    54    50    40    32    91
vowel AW         ticks          67                41

*Pitch 150 from probe 7. At every explicit Pitch the effect's period is ceil(8000/Pitch): one tick shorter than the voice's floor+1 whenever 8000/Pitch is a whole number (100, 160, 200, 250; also 40, 50, 80, 125). The voice keeps floor+1 (41 at Pitch 200). The engine (floor(8026/rate)+1) is therefore one tick long for effects at those Pitches: at Pitch 200 that is 195.8 Hz against the chip's 200.6.

Unresolved: at the default pitch the effect runs 92 ticks, which no rounding of 8000/88 gives. Either the default is not 88 (ceil(8000/87) = 92) or the default period comes from elsewhere. ZZ and ZH after an explicit PITCH 88 (section 6) are too short to settle it (+-1 tick). Next probe: C5 and a long vowel after an explicit PITCH 88, beside the default.

## A tone at next to nothing (section 12, SCP oscillator 1)

0 Hz: silence. 1, 2, 4, 8 Hz: nothing above the background worth measuring. 16 Hz: faint. 32 and 64 Hz: strong low tones
(-7 dBFS). No 1 kHz anywhere. So the 1 kHz harmonic series seen at sound onsets and in A8's pause at Bend 0 and 15 is made by the speech engine's handling of Bend, not by an oscillator sitting at zero.

## Bend (section 10, 30 items, both units agree)

- B2's 1412 Hz tone: 1213, 1257, 1304, 1392, 1547, 1854 Hz at Bend 0, 1, 2, 4, 8, 15: exactly 44 Hz a step with the half step below 5, as the library has it.
- A0's 200 Hz formant (A0 is voiced; its lines are harmonics of a ~69 Hz pulse): bent to 90 Hz (Bend 2) it is at 76-91 Hz, normal; bent to 46 Hz (Bend 1) and 2 Hz (Bend 0) it is not there but a formant appears near 2380 and 2430 Hz, falling about 150 Hz over the part while the bent tone rises 60 Hz. The wrap threshold is between 46 and 90 Hz. The wrapped position is NOT a mirror of slope -1.
- A5 has a tone the library lacks, at about 40 Hz unbent: at Bend 15 it is at 480 Hz and as loud as the main tone (0 dB), at Bend 8 172 Hz, at Bend 4 61 Hz; at Bend 0, 1, 2 (bent below zero) it appears at 2542, 2504, 2463 Hz,
  -12 dB, gliding with the main tone. Slope against the bent frequency about -0.9.
- A8 and C5 at Bend 0-2 show the same kind of lines near 2.0-2.5 kHz; not yet attributed tone by tone.

What to do: fit the wrapped position against the bent frequency over A0, A5, A8, C5 at Bend 0, 1, 2 on both units (a proper fit, not by eye); add A5's 40 Hz tone to its table (it matters at every Bend but 5); then model the wrap.

## Not yet looked at
Sections 2-9 (the earlier questions: Slow/Fast between sounds, reading ahead, the effects' Speed, the nasals' pulse, the loud hisses, the robots, the bed against level, SCP harmonics and mixing products).

Andrew Davie

Quote from: SpiceWare on 02 Nov 2026, 03:03 AMThere's also this limit for attachments, which most of the probes would exceed if I attach both recordings to the same reply:

QuoteRestrictions: maximum total size 60 MB

A php limit, it seems. Forum software limits it to that as a maximum.

Andrew Davie


Andrew Davie

Quote from: Andrew Davie on 02 Nov 2026, 03:20 AM
Quote from: SpiceWare on 02 Nov 2026, 02:59 AM
Quote from: Andrew Davie on 02 Nov 2026, 02:47 AMambiguity of "results"

It seems my usage may be incorrect.

Just for fun here's Claude's take on it...

Both of you are counting different units, and both choices are defensible.

Your logic: the unit is the file. Two files exist, one posted, so "here is the result." Consistent and clear.

Their logic: the unit is the sound. One file holds several recorded sounds, each of which is an outcome, so "here are the results" describes what's inside it. Equally consistent. It's the same as handing someone one PDF and saying "here are the test results."

There's also a third reading: "results" as a loose collective noun for output, the way people say "check out the results" about a single photo of a finished paint job. Common in both British and American English, mildly imprecise, nobody blinks.

With two files and only one posted, "here are the results" implies completeness, so a reader could reasonably assume that's everything. "Here's one of the results" or "here's the result for X" would have avoided that. That's a clarity complaint.




Andrew Davie

Quote from: Andrew Davie on 02 Nov 2026, 03:20 AM
Quote from: SpiceWare on 02 Nov 2026, 02:59 AM
Quote from: Andrew Davie on 02 Nov 2026, 02:47 AMambiguity of "results"

It seems my usage may be incorrect.

Just for fun here's Claude's take on it...

Both of you are counting different units, and both choices are defensible.

Your logic: the unit is the file. Two files exist, one posted, so "here is the result." Consistent and clear.

Their logic: the unit is the sound. One file holds several recorded sounds, each of which is an outcome, so "here are the results" describes what's inside it. Equally consistent. It's the same as handing someone one PDF and saying "here are the test results."

There's also a third reading: "results" as a loose collective noun for output, the way people say "check out the results" about a single photo of a finished paint job. Common in both British and American English, mildly imprecise, nobody blinks.

---

Can't help myself; i was curious is all.
It doesn't matter - the files arrived and all good. TYVM.

SpiceWare

Quote from: Andrew Davie on 02 Nov 2026, 03:24 AMWith two files and only one posted, "here are the results" implies completeness, so a reader could reasonably assume that's everything. "Here's one of the results" or "here's the result for X" would have avoided that. That's a clarity complaint.

From my perspective I did specify the results as being from the "AtariVox+ Enhanced Vectrex" or from the "original Atarivox".

No biggie though, link as I think that phrase is an Americanism.

Andrew Davie

Quote from: SpiceWare on 02 Nov 2026, 03:43 AM
Quote from: Andrew Davie on 02 Nov 2026, 03:24 AMWith two files and only one posted, "here are the results" implies completeness, so a reader could reasonably assume that's everything. "Here's one of the results" or "here's the result for X" would have avoided that. That's a clarity complaint.

From my perspective I did specify the results as being from the "AtariVox+ Enhanced Vectrex" or from the "original Atarivox".

No biggie though, link as I think that phrase is an Americanism.

I regret mentioning it, TBH.