Files
VCCPlusDiagnostics/examples/sound-lab/COVERAGE.md
T

17 KiB
Raw Blame History

Sound Lab coverage and remaining acceptance

This is an initial diagnostic disk and listening toolkit, not completion of the entire Sound Lab specification. The current set contains 114 authored vectors with vector-set CRC 62E5. Adding more vectors will require an explicit vector revision and renewed artifact receipts; existing reports must remain attributable to their original bytes.

Reference: Sound Lab shared CoCo Sound/Speech diagnostic specification, Revision 1, November 17, 2026, repository blob 8b2b303cfce4a3a7e4d1bf377b6a8fca68e62b16: original specification.

How to interpret the coverage

Implemented packet means an authored packet exists and passes the host's grammar/identity checks. Partial means only part of a specified activity is implemented. Deferred means it cannot be accepted using this disk's current cases. None of these labels means Brian heard a sound, that a physical cartridge passed, or that the FPGA implementation is correct.

Vector subcase IDs distinguish individual packets from whole specification chapters. For example, BUF01-SP1 addresses speech slot 1; it does not prove the full BUF01 overwrite-isolation/sharing requirement. ALP01-01 is just pause address 1 within the specification's 64-address ALP01 chapter. The mappings below are authoritative for this initial set.

The seven-case showcase is TXT01, TON01-A, TON04, TON05, NOI01, FUN01-TUNE, MIX01. The guided group has 14 cases. The engineering group and individual case selector expose the other packets. The engineering menu currently still asks for a human response after each case; it is not an unattended sweep. Duration targets of three minutes and 10–15 minutes depend on listener responses and have not been qualified as fixed-duration experiences.

Specification-to-implementation matrix

Specification ID Current implementation / local vector IDs Remaining requirement and acceptance
SET01 Partial. Guest asks target, CoCo model, NTSC/PAL and disk/screen-only mode; user confirms normal speed and configured cartridge. Existing SDK routing token is saved/restored. No floating-status auto-detection or MPI bank changes. Cartridge model/modification, actual MPI slot arrangement, audio path and emulator/core identity are not all collected as structured guest fields. Record these in accompanying run notes. Physical routing/interrupt preservation still needs observation.
RST01 Partial infrastructure. Startup uses FF7D=1 then0; subsequent output has bounded ready polling. No separately recorded cold-reset test, startup status timeline or reset-silence listening result. A click or sustained sound cannot be inferred from a later ready bit.
TXT01 Implemented packet. Exact original greeting, followed by explicit space/CR; replay and HUMAN score supported. Native/emulated execution evidence and real-card listening are separate. Record recognizability; do not substitute modern text-to-speech pronunciation.
TXT02 Partial. TXT02-READ says an authored sentence with words and spelled-out numbers; exact ASCII is in manifest. Minimal pairs, digit-character cases and an actual countdown are not implemented.
TXT03 Partial. TXT01/TXT03-LONG provide different sentence lengths and manual replay. Exact lengths 1/31/62, delayed submission, firmware boundary behavior and maximum-capacity qualification are not tested. Host rejection of oversized packets is not a firmware acceptance test.
ALP01 Implemented packet series. Local ALP01-00–ALP01-63 request all 64 addresses independently, label each and explain addresses 0–4 as silent pauses; final pause and FF included. Group navigation is sequential cases, not an authored small-group lesson. All pronunciation, pause timing and physical listening remain unqualified.
ALP02 Deferred. Individual codes do not make a word-comparison activity. Add original hand-authored words versus equivalent text words, exact code arrays, repeated A/B playback and optional blind transcript reveal.
BUF01 Partial. BUF01-SP0–BUF01-SP7 independently load and execute all eight text slot addresses using unique spoken labels. These cases do not prove retained neighbors, overwrite isolation, repeated playback without reload, or text/allophone sharing. Add one multi-step script that loads all slots, replays them, overwrites one, and replays its neighbors without reloading.
BUF02 Partial. BUF02-SN0–BUF02-SN7 independently load and execute all eight sound slot addresses with distinct pitches. No semantic sound-slot isolation or simultaneous speech-buffer independence proof. Load all first, replay without reload, overwrite one and verify unchanged neighbors and stored speech.
CMD01 Partial. Immediate text, stored text, sound, allophone and direct-register packets exist. Local CMD01 sends C7/CF/00. TIM01 sends 8F. No chained-buffer vectors, stored-register B8–BF/F8–FF execution vectors, stop/retain/replay sequence, reload recovery lesson, or dedicated command-family equivalence assertions. FF boundary/data handling is tested on the host and has REG02/TIM02 guest packets.
TON01 Implemented packet components. Local TON01-A/TON01-B/TON01-C independently exercise A/B/C with period 254 and amplitude 10; TON05 is a three-voice chord. Requires actual execution/listening and comparative measurements. These are three mono voices, not three stereo positions.
TON02 Partial. TON04 steps through periods 428/339/254. Specified 250/500/1000 and engineering boundaries 1/4095 are absent. The channel B packet is TON01-B; it does not stand in for this pitch ladder.
AMP01 Partial. Local AMP01 uses amplitudes 2/5/9/13 at one pitch, followed by explicit silence. Exact ladder 0/1/4/8/12/15 with zero-volume rest and repeat is absent. No linear-loudness or equal-gain assertion is made.
NOI01 Partial. Local NOI01 uses noise A periods 1/8/31; NOI02 exercises retain-previous-amplitude bit. Noise B/C, periods 0/16, percussion lesson and tone-plus-noise mixer cases are absent. Analog noise is not expected to be sample-identical across devices.
ENV01 Partial. ENV01-00–ENV01-15 request every shape with documented tone/envelope ordering and explicit final silence. All use one envelope period and one channel. Short/long comparison, retrigger during playback, two-channel sharing and richer A/B descriptions are absent. ENV01-10 belongs to the guided group; the other 15 shapes are engineering cases.
REG01 Partial. REG01 programs toneA fine/coarse, mixer and fixed amplitude through AF; REG02 includes legal FF register data. Direct envelope/multi-channel/mixer isolation matrix is absent. No unsupported AY register readback is invented.
TIM01 Partial. Local TIM01 sets base 16 and plays one finite tone; TIM02 uses duration FF. No controlled multi-duration/multi-base matrix, precise note-on/off timestamps or repeated-gap distributions. A declared timer baseline is required between cases; coarse guest ticks are not calibrated audio timing.
STS01 Partial infrastructure. Guest records full status before/after, transmitted-byte count, elapsed VSYNC ticks and whether a wait-mask busy state was observed. No continuous transition timeline, startup/decay latency distribution or calibrated polling resolution. A missed pulse is unresolved, not evidence of no pulse. Status 40/20 meanings remain source/hardware qualification work.
MIX01 Partial. TXT01, tone cases and local MIX01 offer speech alone, sound alone and a held tone under an authored phrase; bounded global cleanup stops both. No independent speech-stop versus sound-stop listening sequence or structured clipping/dropout comparison. Held-tone cleanup must be confirmed in execution.
MUX01 Deferred. Routing is set on entry and restored on exit. No mid-play cartridge mux/speaker gating exercise, inaudible-progress comparison or normal CoCo SOUND/DAC regression. Entry/exit restoration is not equivalent to this test.
RST02 Partial safety infrastructure only. Timeout/interrupted packets reset before recovery. No deliberate active-tone, active-speech, combined-play or incomplete-load reset cases with independently recorded recovery and explicit buffer reinitialization.
REP01 Deferred. The user can repeat a selected case. No numbered repeated speech sequence covering cumulative 4096 speech samples, no stale-ring/frozen-speech detector. A guest cannot read internal speech sample counts.
REP02 Deferred. Sequential engineering cases have deterministic order and bounded waits. No timed 60-second/5-minute alternating stress script, unattended mode or qualified drift measurement.
FUN01 Partial experience. Local FUN01-TUNE plays an original finite tune with note-off; greeting and other labels are separate cases. No interactive musical keyboard or spoken note labels. Local ID does not satisfy the whole proposed activity.
FUN02 Partial ingredients. TON05, NOI01, FUN01-TUNE and MIX01 supply chord/noise/tune/spaceport-message components. No integrated countdown, layered engine, launch motif, radio message, landing chime and final quiet timeline. Local MIX01 is a short finale, not the full scene.
ORC01 Deferred optional extension. No FF7A/FF7B packets are sent. Separate Orchestra-90 profile and left/right tests may follow. Not applicable to SSC-only acceptance; SSC is mono.
EXT01 Deferred MiSTer integration. No extended-video-palette operations are performed. FPGA team must test its bus ownership/graphics/ordinary sound/Orchestra interactions. Not applicable to a plain physical SSC.

Other specification requirements

The build produces ordinary 6809 .BIN, Disk BASIC .BAS, 35-track raw .DSK, SHA-256 artifact receipts and the exact vector manifest. The same guest bytes can be copied to each target; platform selection only changes report metadata. The guest has no network, OS-9, GUI debugger or emulator-private API dependency.

The guest polls once per actual bus read, waits before each byte, uses a VSYNC deadline plus an independent iteration limit, and resets after partial-packet failure. These paths exist in source. Their presence is distinct from proving all timeout, panic, IRQ-mask, PAL, counter-wrap and recovery behavior on hardware. The test never selects a faster physical CoCo clock. A speed comparison is not part of this initial disk.

Disk reports use unique names, CRLF, per-case flushes, explicit partial/complete footers, record counts and CRCs. Screen-only mode and RAM paging are present. The bounded RAM report can span multiple report files; preserve all files for a run session. The host extractor/parser compares build/vector identities and keeps AUTO and HUMAN outcomes separate. The initial guest records observations, not automatic hardware PASS. Complete physical disk-full/write-protect/retry, interrupted-report, no-disk and photograph usability checks remain acceptance work. Host synthetic filesystem fixtures do not prove a real drive's behavior.

The optional WAV analyzer accepts manually marked intervals, measures basic PCM levels, isolated-tone crossings, clipping and repeated-case differences. Automatic case-marker alignment, spectrum comparison, calibrated capture-clock tolerances and full audio timeline integration remain deferred. WAV generation from an emulator is not a physical recording, and a text report is not audio.

The specification's comprehensive chip-internal suite and timestamped PIC/AY/SP tracing are a separate deferred engineering effort. The initial native runner can exercise actual emulator DOS boot, selected guest interactions, report disk writes, signal presence and absent-card recovery. It does not implement all timer/IRQ/DDR/cycle-debt, ROM-loader, ALD/DRQ, envelope/noise, coefficient/filter saturation tests or negative-control fault injections requested in that section. Existing unrelated SSC tests must be cited individually before reusing their evidence; they are not silently counted as completion of this disk's spec.

Verification layers and evidence limits

  1. Source/packet checks: six vector host tests passed during initial authoring, including CCITT-FALSE reference value, malformed packets, FF operand handling, all slots/codes/shapes, deterministic outputs and selected fixed vector CRCs. The source catalog records a hashed OCR/typeset Tandy reproduction whose command-table pixels were inspected. Original photographic scan comparison remains pending. Successful validation proves the declared packet structure and identity, not firmware execution or audible fidelity.
  2. Native/emulated checks: use the coordinator's final build/test receipt for exact artifact hashes, source revision and actual results. A compiled runner or generated disk alone is not a passed test. Signal detection must stay separate from HUMAN listening. Any emulator findings describe the current comparison baseline; they do not revise physical expectations to fit it.
  3. Physical acceptance: not performed by this coverage review. Brian must boot this exact disk on a real CoCo, listen/repeat, save/reload reports and verify safe recovery. Physical Intel/macOS testing is unrelated to the CoCo cartridge's analog behavior.
  4. FPGA acceptance: pending research-core availability and its pre-install checks. Use the same disk/vector identities, identify stock versus research cores, and compare observations with the physical baseline. VCC-only success cannot stand in for real-card or MiSTer acceptance.

No speech/sound engine, SDK or application implementation change is authorized by this document. The emulator baseline remains unchanged while FPGA work is finished and recommendations are reviewed.

Next checks for Brian and the FPGA project

  1. Keep an unchanged master disk and manifest. Use a fresh writable copy per platform; verify its starting SHA-256 before first use. After testing, retain the changed disk and all SNDxxxx.TXT files rather than comparing its modified hash to the pristine artifact hash.
  2. Record CoCo model, NTSC/PAL, normal speed,26-3144/3144A/unknown cartridge, known modifications, MPI/disk slots, speaker/audio path and emulator or FPGA build identity. Do not guess physical ROM hashes. No hardware modification or cartridge hot-swap is required.
  3. Boot with DOS on VCC Plus and a real CoCo. Run Meet the Robot at low volume. Score each case, repeat TXT01, mark an intentionally uncertain result, add a note and confirm each sound stops. The current seven cases are listed above; missing full-spec scenes must not be marked complete.
  4. Quit normally, relaunch and run again. Verify a new report filename is used. Safely copy the disk, extract reports, convert them using the matching build manifest, and verify their checksum/count/footer and build/vector identities. Keep actual VCC and physical reports together with any photographs/recordings.
  5. Select individual REG01, REG02, TIM02, representative envelope cases and ALP01-00–ALP01-04. Confirm legal FF operands do not terminate early, held tones end at cleanup, pause cases are intentionally quiet, and no physical timeout threshold is inferred from the emulator's observed count.
  6. Exercise X during an audible wait and confirm a bounded return to menu; repeat TXT01 afterward. With no SSC configured, run screen-only mode and confirm bounded handling/no false detected-card claim. Preserve any timeout or interrupted report instead of treating it as a pass.
  7. On disposable media only, separately test screen-only, write-protected or full disk, retry-save, RAM paging and interrupted-run report extraction. Confirm the app never formats a disk or overwrites an earlier run. A full RAM report may need a new file/run; save or photograph before clearing.
  8. If recording audio, use fixed gain and preferably 48 kHz/16-bit PCM with lead-in and trailing silence. Mark exact intervals manually, retain the raw WAV and capture notes, and compare measured changes without demanding identical analog waveforms or absolute loudness. Record human judgments independently.
  9. Give the FPGA team the same pristine disk, .BIN/.BAS, full manifest, parser/analyzer, this coverage map and actual physical reports. Run first on the separately identified research core after its own build/pre-install checks. Report discrepancies without modifying the emulator baseline.
  10. Expand deferred tests deliberately: first buffer isolation/sharing and reset recovery; then exact tone/amplitude/noise/timing matrices; then mux controls, stress runs and chip-internal negative controls. Address the ID mapping explicitly in that next vector revision, preserving subcase suffixes so an individual packet cannot be mistaken for a complete specification chapter.