Module voice 

Voice capture: microphone-to-WAV pipeline.

The library half of omni-dev voice capture. The CLI entry point lives in crate::cli::voice. This module is intentionally CLI-free so the audio pipeline (source → mixdown → resample → idle-detect → trim → write) can be unit-tested against fixture WAVs without a real microphone.

The AudioSource trait in audio is the test seam: production code uses audio::CpalAudioSource, tests use audio::FileAudioSource. See ADR-0031 for the rationale.

Re-exports

pub use audio::AudioSource;

pub use audio::CpalAudioSource;

pub use audio::FileAudioSource;

pub use capture::install_ctrl_c_handler;

pub use capture::run_capture;

pub use capture::CaptureOpts;

pub use capture::CaptureSummary;

pub use capture::TerminationReason;

pub use clock::Clock;

pub use clock::FixedClock;

pub use clock::SystemClock;

pub use det::CountingUlidRng;

pub use det::SystemUlidRng;

pub use det::UlidRng;

pub use factory::create_default_transcriber;

pub use factory::VoiceOpts;

pub use paths::captures_dir;

pub use paths::omni_dev_voice_root;

pub use paths::speaker_file;

pub use paths::speakers_dir;

pub use render::detect_format;

pub use render::render_jsonl;

pub use render::render_markdown;

pub use render::OutputFormat;

pub use speaker::cosine;

pub use speaker::l2_normalise;

pub use speaker::EnrolledSpeaker;

pub use speaker::WespeakerEmbedder;

pub use speaker::MIN_EMBED_SAMPLES;

pub use transcriber::AudioChunk;

pub use transcriber::AudioInput;

pub use transcriber::EndpointKind;

pub use transcriber::EventId;

pub use transcriber::EventStream;

pub use transcriber::SpeakerId;

pub use transcriber::Transcriber;

pub use transcriber::TranscriptEvent;

pub use transcriber::VecAudioInput;

pub use transcriber::Word;

Modules

audio

Audio source abstraction.

backends

Transcriber backends.

capture

End-to-end capture pipeline orchestrator.

clock

Pluggable wall clock for deterministic timestamps in tests.

det

Pluggable RNG for EventId (ULID) generation.

events

Reflection event schema (the events.jsonl contract from #799).

factory

Backend factory for crate::voice::Transcriber.

features

Kaldi-style FBANK (log-mel filterbank) feature extraction.

idle

Idle (silence) detection and trailing-silence trimming.

models

Model storage convention and path resolution.

paths

User-state directory helpers for the voice subsystem.

reconcile

Pure reconciliation of events.jsonl into materialised markdown.

reflect

voice reflect — transcript-to-events Claude consumer.

render

Rendering helpers for crate::voice::TranscriptEvent streams.

review

voice review driver — wraps the pure crate::voice::reconcile function with the I/O the CLI needs.

session

~/.omni-dev/voice/<id>/ session directory I/O.

speaker

Speaker embedding via tract-onnx + wespeaker, plus the persisted enrolled-speaker JSON shape.

transcriber

Transcriber trait and event types per issues #799 and #801.

wav

Mixdown, resampling, and WAV writing.