Struct SynthesisStageTimings 

pub struct SynthesisStageTimings {
    pub speaker_encode: Duration,
    pub tokenize: Duration,
    pub prefill_build: Duration,
    pub codec_rollout: Duration,
    pub vocoder_decode: Duration,
    pub post: Duration,
    pub pipeline_overlap: Duration,
    pub first_frame_latency: Duration,
    pub codec_rollout_detail: CodecRolloutSubTimings,
    pub generated_samples: usize,
    pub sample_rate_hz: u32,
}

Per-stage wall times for a single crate::Qwen3TtsEngine::synthesize_with_profile call.

When pipelining is disabled, stages are sequential and their sum approximates end-to-end latency. With pipelining (vocoder_chunk_size > 0), codec_rollout and vocoder_decode overlap; use [total()] which accounts for this via pipeline_overlap.

Fields

speaker_encode: Duration

Reference WAV → speaker embedding (zero if not used).

tokenize: Duration

Text → token ids (includes voice-clone ref+target path when applicable).

prefill_build: Duration

build_prefill_inputs (text projection + codec input layout).

codec_rollout: Duration

rollout_codec_frames_kv (talker prefill + autoregressive codec steps).

vocoder_decode: Duration

Vocoder decode to PCM.

post: Duration

Flatten codec rows + optional ICL prefix trim on PCM.

pipeline_overlap: Duration

Time saved by overlapping transformer and vocoder (pipelining). Zero when pipelining is disabled.

first_frame_latency: Duration

Time from synthesis start to first generated codec frame (includes speaker encode + tokenize + prefill + first frame prediction). This is the theoretical minimum latency for streaming playback.

codec_rollout_detail: CodecRolloutSubTimings

Sub-component breakdown within the codec_rollout stage.

generated_samples: usize

Number of PCM samples in the final output (after prefix trim).

sample_rate_hz: u32

Output sample rate (Hz).

Implementations

impl SynthesisStageTimings

pub fn total(&self) -> Duration

pub fn average(samples: &[Self]) -> Option<Self>

Arithmetic mean of each stage (wall clock). Empty slice returns None.

pub fn format_table(&self) -> String

Multi-line table: stage name, milliseconds, % of total.

Trait Implementations

impl Clone for SynthesisStageTimings

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 SynthesisStageTimings

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

Formats the value using the given formatter.

impl Default for SynthesisStageTimings

fn default() -> Self

Returns the “default value” for a type.

impl Eq for SynthesisStageTimings

impl PartialEq for SynthesisStageTimings

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

Equality operator ==.

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

Inequality operator !=.

impl StructuralPartialEq for SynthesisStageTimings