Struct Config 

pub struct Config {
    pub models_dir: Option<PathBuf>,
    pub asr_model: ModelSpec,
    pub normalizer: Option<ModelSpec>,
    pub normalize_options: NormalizeOptions,
    pub language: Option<String>,
    pub max_capture: Duration,
    pub silence_floor: f32,
    pub diagnostics_dir: Option<PathBuf>,
    pub digest_cache_dir: Option<PathBuf>,
}

Configuration for one dictation engine.

Construct via ConfigBuilder rather than populating directly. Values are validated where they are used, never here: an unreadable models directory surfaces from crate::prepare, not from ConfigBuilder::build.

Fields

models_dir: Option<PathBuf>

Directory holding model files. None resolves to a subdirectory of the platform cache directory.

asr_model: ModelSpec

Speech recognition model.

normalizer: Option<ModelSpec>

Model that rewrites raw recognition output into clean text. None leaves the recognition output as-is and fetches nothing for it.

When set, the engine loads it beside the recognition model and rewrites every transcript: crate::Transcript::text carries the result and asr the recognised text it came from. When None the two are equal, which is a supported configuration rather than a degraded one.

normalize_options: NormalizeOptions

How the normalizer rewrites text, when one is configured.

styling is the axis a person picks per recording; k and ngram are decoder tuning and do not belong in a user-facing picker beside it. Overridable per call via crate::Engine::transcribe_with, so this is the default rather than a fixed choice.

language: Option<String>

Spoken language hint. None autodetects.

max_capture: Duration

Upper bound on a single capture. A capture nobody stops ends itself here rather than holding the input device indefinitely.

Costs memory eagerly: the whole cap is reserved when recording starts, at 4 bytes a sample, so the 30 min default reserves about 110 MiB and an hour reserves 219 MiB, whatever the utterance turns out to be. Reserved rather than grown because the audio callback must not allocate. Clamped to an hour.

silence_floor: f32

Peak input level, in dBFS, below which a capture counts as silence rather than as an empty transcript.

diagnostics_dir: Option<PathBuf>

Directory the crate writes per-take diagnostics into: the capture audio and every transcription stage, best-effort, under crate::diagnostics. None writes nothing - diagnostics are the host’s choice of directory, never a requirement.

digest_cache_dir: Option<PathBuf>

Directory the crate records verified model digests in, keyed on each file’s size and modification time, so a file unchanged since it was last verified is not read end to end again.

None disables the cache and hashes every time: a caller with no machine-local directory of its own to put it in keeps the old cost rather than having one invented for it.

Trait Implementations

impl Clone for Config

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Config

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for Config

fn default() -> Self

Returns the “default value” for a type.