Documentation update following the Android audio spike pass.
Decision register (v0.9.3 → v0.9.4):
- DEC-011.1 promoted from
'Accepted (desktop: cpal) / Deferred (mobile)'
to
'Accepted (desktop: cpal; Android: cpal-on-Oboe) / Deferred (iOS)'.
- Evidence pointer added: poc/audio-capture-playback-android-spike/
VERIFICATION.md.
PoC plan (v0.2.0 → v0.3.0):
- Audio row promoted from PARTIAL PASS to PASS.
- All six PoC plan entries are now PASS.
PoC results summary (v0.1.0 → v0.2.0):
- Audio row collapsed into one PASS spanning both spikes.
- RISK-PoC-001 narrowed from 'mobile audio' to 'iOS audio only'.
- Toolchain table expanded with Android NDK, cargo-ndk, AGP/
Gradle/Kotlin, jni/ndk-context/android_logger, and the test
device.
Cross-spike pointers updated:
- poc/audio-capture-playback-spike/VERIFICATION.md result and
follow-up sections updated to reference the Android spike.
- poc/README.md status table lists both audio spike directories.
CHANGELOG updated under [Unreleased].
4.7 KiB
4.7 KiB
Verification record — audio-capture-playback-spike
Result
PASS (desktop half). The Linux/PipeWire desktop half of the PoC plan §2 audio exit criterion is fully met by this spike.
The mobile half ("at least one mobile target") is closed by
poc/audio-capture-playback-android-spike
on a physical Android device — see that directory's VERIFICATION.md.
Combined: the audio PoC is fully retired.
Environment
| Field | Value |
|---|---|
| Date | 2026-05-13 |
| Host OS | Linux (Arch, kernel 7.0.5-arch1-1, x86_64) |
| Rust toolchain | stable 1.95.0 |
cpal |
0.16 |
hound (WAV I/O) |
3.5 |
| Audio server | PipeWire 1.6.4 (with the pcm_pipewire ALSA plugin module) |
| Default input | auto_null (system has no physical microphone bound to this session) |
| Default output | auto_null (PipeWire null sink) |
| Local config | ~/.asoundrc aliases pcm.!default and ctl.!default to pipewire |
Reproduction
cargo test
cargo run --bin audio-capture-playback-cli -- roundtrip
Test run
running 2 tests
test synth_produces_valid_wav_with_expected_frame_count ... ok
test synth_then_playback_does_not_error_when_output_device_present ... ok
test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured;
0 filtered out; finished in 0.06s
CLI round-trip
INFO spike: opening capture stream device="default" sample_rate=44100
channels=2 sample_format=F32
WARN spike: capture unavailable; falling back to synth
error=stream build error: A backend-specific error has
occurred: ALSA function 'snd_pcm_hw_params' failed with
error 'No such file or directory (2)'
INFO spike: synthesised fallback frames=24000 sample_rate=48000
INFO spike: preparing playback device="default" file_sr=48000
file_channels=1 samples=24000
INFO spike: playback finished consumed=24000 total=24000
round-trip complete (played from /tmp/chanora_audio_spike_synth.wav)
What this evidences
- cpal successfully opened the default input device (negotiated
44.1 kHz / 2-channel / F32 against PipeWire's
auto_null). - The capture stream build then failed because the null source
does not expose
snd_pcm_hw_paramsat that configuration. This is a real platform-audio failure mode; the typedAudioError::BuildStreamwas returned and surfaced cleanly. This is the same shape production code will see on misconfigured user hosts. - The fallback synth wrote a valid 24,000-frame, 48 kHz, mono, 16-bit WAV.
- cpal opened the default output device, negotiated the
matching config, and the playback callback drained the WAV
completely:
consumed=24000 total=24000. Process exited 0.
The full path Rust → cpal → ALSA → pcm_pipewire → PipeWire →
auto_null ran end-to-end, with PCM frames produced by Rust
consumed by the platform audio stack.
Coverage matrix
| PoC plan requirement | Status |
|---|---|
| Desktop capture path | PASS (stream opens against the system audio stack; typed error returned when device cannot satisfy hw_params). |
| Desktop playback path | PASS (24,000 frames driven to completion). |
| Mobile capture path | NOT VERIFIED. |
| Mobile playback path | NOT VERIFIED. |
| Round-trip flow | PASS — capture-attempt → fallback → playback. |
| Typed error model | PASS — AudioError arms returned cleanly. |
| Cross-platform crate choice (DEC-011 alignment) | PARTIAL — cpal is a platform-native abstraction, but only the Linux back-end exercised here. |
What this spike does NOT validate
- Real microphone capture (this host has none).
- Bit-exact capture-to-playback signal integrity.
- Latency.
- Hot-plug / device-change events.
- Permission flows (mobile microphone consent, Linux pulse permission prompts).
- Behaviour under sample-rate or channel-count mismatch between file and device (the PoC chose the file's rate; production code will need a real resampler).
- DSP (HPF / NS / AEC / AGC).
- Opus encode / decode.
- Jitter buffer / mixer.
- Mobile (Android Oboe / iOS AVAudioEngine) — explicit gap.
Recommended follow-up
- Promote
cpalselection to a recorded product decision (DEC-011 sub-decision) — or pin a different crate after the mobile spike. Done on 2026-05-13 as DEC-011.1 indocs/governance/product-decision-register.md. - Spike Android Oboe path on an emulator or real device.
Done on 2026-05-13 — see
poc/audio-capture-playback-android-spike/VERIFICATION.md. - Spike iOS AVAudioEngine path on a real device. Still deferred; requires macOS + Xcode hardware.
- Only after the iOS half is closed (or explicitly waived):
promote the typed
AudioError+ capture/playback shapes intochanora_audio.