Struct SincResampler

pub struct SincResampler { /* private fields */ }

Expand description

One-channel f64 arbitrary-ratio streaming converter, rate ratio 1/16..16. Preparation allocates a 1025-phase Blackman-windowed sinc table and history. Processing interpolates table phases and allocates nothing. Filter delay is (taps-1)/2 input samples; initial history is zero. This is a fixed quality profile, not a spectral rejection or callback deadline guarantee. Fixed rates use Self::new; Self::new_integer_rates selects an exact integer schedule; Self::new_variable prepares bounded live ppm adjustments with a conservative fixed cutoff and retained stream phase.

Implementations§

§

impl SincResampler

pub fn new( input_rate: f64, output_rate: f64, taps: usize, ) -> Result<Self, ResamplingError>

Finite positive rates and even tap count 16..256. Control-thread work. Rates are expressed in the same units; no hardware/format conversion.

pub fn new_integer_rates( input_rate: u32, output_rate: u32, taps: usize, ) -> Result<Self, ResamplingError>

Prepare a caller-selected exact integer-rate schedule. Positive u32 rates, ratio 1/16..16 and even 16..256 taps; coefficient/history preparation matches the fixed profile. Scheduling uses a bounded signed relative numerator, without duration-dependent floating recurrence. Only phase interpolation uses f64; numerical PCM can differ from new. This does not improve the spectral profile or establish device/callback deadlines. Nonzero live ppm adjustments reject; finite streams still require caller-supplied padding.

pub fn integer_position( &self, output_frame: u64, ) -> Result<IntegerResamplingPosition, ResamplingError>

Compute the exact next-output input coordinate without changing state. Integer profiles only; u128 input count includes explicit caller padding. No allocation, I/O, source validation or finite-file tail policy is implied.

pub fn prime_integer_at( &mut self, output_frame: u64, chronological_history: &[f64], ) -> Result<IntegerResamplingPosition, ResamplingError>

Prime an integer stream so its next sample is the specified output frame. Off-audio source-window preparation supplies exactly taps finite samples oldest-to-newest, ending at input coordinate C-1 from integer_position. Leading/trailing zeros and finite-file bounds belong to the caller. Invalid mode/history preserves all state. This operation allocates nothing and does not read, seek, authenticate source history or change coefficients/rates.

pub fn new_variable( input_rate: f64, output_rate: f64, taps: usize, maximum_adjustment_ppm: f64, ) -> Result<Self, ResamplingError>

Prepare a live-adjustable clock-drift profile. Maximum adjustment must be finite in (0,10000] ppm (one percent); the entire permitted input/ output step range must stay within 1/16..16. Filter cutoff uses the largest permitted step, avoiding live coefficient allocation. This is not an automatic drift estimator or a spectral/deadline guarantee.

pub fn validate_variable( input_rate: f64, output_rate: f64, taps: usize, maximum_adjustment_ppm: f64, ) -> Result<(), ResamplingError>

Check live-adjustable preparation bounds without allocating coefficient storage or mutating a stream. This does not prepare a converter or prove native device support. Rates are finite positive values in matching units.

pub fn set_rate_adjustment_ppm( &mut self, ppm: f64, ) -> Result<(), ResamplingError>

Change future output spacing without resetting history or pending phase. Positive ppm consumes input faster per output sample. The next output’s already scheduled position is retained; new spacing applies afterward. Reject before mutation. Fixed constructor accepts only zero adjustment. No smoothing/estimation is implicit; this operation allocates nothing.

pub fn rate_step(&self) -> f64

Current input samples per output sample, including live adjustment.

pub fn maximum_adjustment_ppm(&self) -> f64

pub fn latency_input_samples(&self) -> f64

pub fn reset(&mut self)

Clear stream state and zero history without reallocating coefficients. Retain the current rate adjustment; set zero explicitly to return nominal.

pub fn process( &mut self, input: &[f64], output: &mut [f64], ) -> ResamplingProgress

Consume/produce bounded prefixes, retaining pending output internally. Zero output capacity consumes nothing. Zero input may drain pending output. Output beyond produced is untouched. To flush a finite stream, supply explicit trailing zeros; this method does not invent input or final padding. Nonfinite audio propagates. Fixed construction retains its rate; the variable profile accepts only its prepared bounded ppm adjustments.