Struct Stages 

pub struct Stages {
    pub model_load: Duration,
    pub resample: Duration,
    pub mel: Duration,
    pub encode: Duration,
    pub decode: Duration,
    pub normalize: Option<Duration>,
    pub audio: Duration,
}

Where the time went in one transcription.

Which stage moved is what separates “our code got slower” from “the model changed”.

Fields

model_load: Duration

Loading weights. Zero on every run but the first, which is why it is separate: a cold start is not a regression.

resample: Duration

Turning the source into the buffer the model reads.

mel: Duration

Mel spectrogram, as reported by the recognition runtime.

encode: Duration

Encoder pass, as reported by the recognition runtime.

decode: Duration

Decoder pass, as reported by the recognition runtime.

normalize: Option<Duration>

Rewriting recognition output into clean text. None when normalization did not happen - either none was configured, or it failed and the raw text was returned instead. A caller knows which, because it knows what it configured.

audio: Duration

How much audio this run consumed, so a caller can derive a realtime factor without knowing where the audio came from.

Trait Implementations

impl Clone for Stages

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 Stages

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

Formats the value using the given formatter.

impl Default for Stages

fn default() -> Self

Returns the “default value” for a type.

impl PartialEq for Stages

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for Stages