Conversation
The Adapter and AdapterMut traits no longer carry a lifetime parameter, so all uses become Adapter<T> / AdapterMut<T>.
Indexing gains Default, new(), and chainable setters named like its fields. SincInterpolationParameters gains new(sinc_len, window) which derives f_cutoff via calculate_cutoff, with setters that keep f_cutoff consistent. Examples and the README usage block now use the fluent forms.
The synchronous Fft resampler now exposes the anti-aliasing window: new is simplified (drops sub_chunks, picking one automatically, default window) and a new new_custom exposes both sub_chunks and the window function. SincInterpolationParameters::f_cutoff becomes an Option<f32>. None (the default) derives the cutoff from sinc_len and window via calculate_cutoff, resolved after sinc_len is rounded so it matches the filter actually built. Some(value) overrides.
Derive Clone, Copy and PartialEq for ResampleError and ResamplerConstructionError so downstream code can compare, store and assert on them, and mark both #[non_exhaustive] so future variants are not breaking. Mark the chainable Indexing and SincInterpolationParameters setters #[must_use] to catch calls whose result is accidentally discarded.
Defaults to new(256, BlackmanHarris2): sinc_len 256, automatic cutoff, oversampling_factor 128, Cubic interpolation. Enables struct-update syntax and parity with Indexing.
Add Resampler::process_all, an allocating one-shot method for resampling a whole clip: it resets the resampler, trims the startup delay, and returns an InterleavedOwned with exactly the resampled frames. This gives the correct home for one-shot whole-clip resampling, which was previously done by oversizing the resampler and calling process once (leaving the startup delay untrimmed). Change process to take Option<&Indexing> instead of separate input_offset and active_channels_mask args, matching process_into_buffer. This adds partial_len support for a short final chunk (previously callers padded by hand) and makes the common call process(&input, None). output_offset is ignored since the output buffer is freshly allocated. Sharpen the process docs to point whole-clip users at process_all.
Derive PartialEq, Eq and Hash for the configuration enums (WindowFunction, SincInterpolationType, PolynomialDegree, FixedSync, FixedAsync) so they can be compared and used as map keys. Add a 'Migrating from 3.x to 4.0' section to the README with before/after examples for the breaking changes, and rework the 'Resampling a given audio clip' section to point at process_all and warn against the one-shot misuse.
There was a problem hiding this comment.
Pull request overview
This PR performs a major-version bump of rubato to 4.0.0 by upgrading to audioadapter 4.0 (removing Adapter/AdapterMut lifetimes) and introducing several API ergonomics improvements for constructing/configuring resamplers and resampling whole clips.
Changes:
- Upgrade to
audioadapter4.0 and update public APIs accordingly (trait object signatures, examples, benches). - Add fluent constructors /
DefaultforIndexingandSincInterpolationParameters, and makeSincInterpolationParameters::f_cutoffautomatic-by-default viaOption<f32>. - Improve FFT resampler construction (
Fft::newdefault behavior,Fft::new_customfor expert control) and addResampler::process_allfor one-shot clip resampling.
Reviewed changes
Copilot reviewed 18 out of 20 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| src/windows.rs | Derive PartialEq/Eq/Hash for WindowFunction to support comparisons/keys. |
| src/synchro.rs | FFT sync resampler constructor/API updates (window selection, new vs new_custom) and audioadapter 4.0 signature updates. |
| src/lib.rs | Adds Indexing builder + Default, updates process signature, and introduces allocating process_all. |
| src/error.rs | Makes error enums #[non_exhaustive] and adds derives (Clone/Copy/PartialEq). |
| src/asynchro.rs | Updates async resampler signatures for audioadapter 4.0 and derives for FixedAsync. |
| src/asynchro_sinc.rs | Adds SincInterpolationParameters builder + Default, makes f_cutoff optional with automatic cutoff resolution. |
| src/asynchro_fast.rs | Derives PartialEq/Eq/Hash for PolynomialDegree and updates adapter mut signatures. |
| README.md | Updates documentation for new APIs and adds a 3.x → 4.0 migration guide. |
| examples/process_i16.rs | Updates example code to new parameter builders and audioadapter 4.0 API. |
| examples/process_f64.rs | Updates example code to new parameter builders and FFT constructor changes. |
| examples/process_all_f64.rs | Updates example code to new parameter builders and FFT constructor changes. |
| examples/polyfixedin_ramp64.rs | Updates example code to use Indexing::new() builder. |
| examples/fixedout_ramp64.rs | Updates example code to new parameter builders and Indexing::new() builder. |
| Cargo.toml | Bumps crate version to 4.0.0 and upgrades audioadapter dependencies to 4.0. |
| benches/resamplers.rs | Updates benches for FFT constructor changes and window selection. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
update_mask now checks the provided mask length and returns WrongNumberOfMaskChannels, instead of letting copy_from_slice panic on a mismatch. Both process_into_buffer call sites propagate the error. Remove the now-dead mask-length check from validate_buffers: it only ever saw the internal channel_mask (always the right length), so it could never fire.
Move the capability-specific methods off the Resampler trait: set_resample_ratio and set_resample_ratio_relative go to a new Adjustable trait, set_chunk_size to a new Resizable trait. Both extend Resampler. Resampler gains as_adjustable() and as_resizable() to recover the capabilities from a trait object, defaulting to None and overridden to Some(self) by Async. This turns the runtime SyncNotAdjustable / ChunkSizeNotAdjustable errors into compile-time capabilities, so both error variants are removed. Resampler stays the universal trait for mixed Box<dyn Resampler> collections. Update the examples and tests to bring the traits into scope, and document the change in the changelog and migration guide.
… exact ratio bounds - Make `Resampler::as_adjustable`/`as_resizable` required rather than defaulting to `None`, so implementing `Adjustable`/`Resizable` and advertising it through a trait object cannot silently drift apart. `Fft` now declares both as `None` explicitly. - Add `Resampler::is_adjustable`/`is_resizable` so the capability can be queried through a shared `&dyn Resampler` without exclusive access. - Check the relative ratio directly in `set_resample_ratio_relative` instead of round-tripping through `(original * rel) / original`, which could round the exact bounds out of range and wrongly reject them. Adds a regression test. - Fix the stale `Async` doc that referenced `set_chunk_size()` without noting the `Resizable` trait. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The crate-level docs include README.md, so its headline example ran as a doctest. It uses `Fft`/`FixedSync`, which only exist with the default `fft_resampler` feature, so `cargo test --no-default-features` failed to compile it. Mark the block `ignore` (matching the other README snippets) and note the feature requirement. The Fft processing flow is still compile-tested by the `process_f64` example, which gates the import on the feature. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Split Resampler into capability traits Adjustable and Resizable
Slip matches two almost-equal sample rates by occasionally inserting or dropping a frame, hidden by a short smootherstep crossfade, instead of running a full resampler. It adds no delay and no high-frequency roll-off, and its ratio is meant to be adjusted at runtime through Adjustable by a feedback loop. Several frames can be slipped per chunk (up to the crossfade spacing), so it can track up to about 10% drift.
Bind FADE to a local before asserting so the checks run on a runtime value rather than a constant expression.
The hardened capability traits added `Resampler::is_adjustable`/`is_resizable` so the capability can be queried through a shared `&dyn Resampler`. `Slip` is both `Adjustable` and `Resizable`, so it must override both to return `true`; otherwise the defaulted `false` would under-report its capabilities through a trait object. Extend the `capability_queries` test to cover `Slip`. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Companion to process_all_f64: instead of converting between two fixed rates, this drives any of the adjustable resamplers (Async sinc/poly and Slip, fixed input or output) at a nominal 1:1 ratio and applies a small user-selected rate offset given in ppm. This is the clock-drift / rate-matching case. The offset is applied through `Resampler::as_adjustable`, so one code path handles every adjustable type without knowing the concrete resampler. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Replace the fixed 8-frame smootherstep crossfade with a linear ramp whose length adapts to the chunk size, up to 128 frames. Listening tests on sustained pure tones (the worst case) showed the slip artefact is governed by the peak retiming rate of the crossfade: its spectral spread scales with 1/length, so a longer fade keeps the disturbance narrow and masked by the signal. Any velocity shaping that peaks (smootherstep is 1.875x steeper in the middle than its average) or concentrates the retiming (end-loaded curves, a hard cut) only widens it. A linear ramp is the minimum-peak and minimum-energy monotonic fade, so the old 8-frame smootherstep was in practice barely better than a hard cut. - Drop the FADE compile-time table; compute the linear weight inline. - Replace the CROSSFADE_LEN constant with MAX_CROSSFADE_LEN (128) plus crossfade_len_for(chunk), stored per instance and recomputed on resize/reset. Large chunks get the full 128-frame fade; small chunks shrink it instead of being rejected, lowering the minimum chunk size from 18 to 4. - Update the docs to match, and adjust tests for the narrower sustainable ratio range at the longer fade. Also add examples/gen_test_tones.py, the stepped-tone generator used to audition the crossfade. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Address two review findings: - Slip::process_into_buffer copied `input_len`/`output_len` frames from the input buffer, but validate_buffers only guarantees `frames_to_read` frames are present. On a short final chunk (Indexing::partial_len set) that read past the validated region. Copy only `frames_to_read` frames; the remainder of the scratch is already zero-padded. - The adjust_ratio_f64 example read samples with Read::read, which may return a short read and decode a corrupt trailing sample from leftover bytes. Use read_exact and stop on UnexpectedEof. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add Slip resampler for cheap clock-drift rate matching
Switch the main example from the feature-gated `Fft` resampler to the always-available `Async` polynomial resampler so the doctest can run unconditionally (drops the `rust,ignore`). Restructure it around reusable per-chunk buffers and a general loop that works for a file or a live stream. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add a "Choosing a resampler" section that maps common tasks (offline file conversion, drift-tracking streams, low-CPU/voice, pitch sweeps, clock-drift matching, VU meters) to the appropriate resampler and constructor, including the new Slip clutch. Also fix several pre-existing README issues: stale docs.rs links pointing at audioadapter 2.0.0 instead of 4.0.0, typos, broken grammar, and a rustdoc intra-doc link that rendered as literal text on GitHub. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Major release: update to
audioadapter4.0 plus a round of API ergonomics improvements.Changes
audioadapter4.0 (theAdapter/AdapterMutlifetime is gone)DefaultforIndexingandSincInterpolationParametersSincInterpolationParameters::f_cutoffis nowOption<f32>(automatic cutoff by default)Fftresampler can select the anti-aliasing window (newsimplified, newnew_custom)process_allfor one-shot whole-clip resampling;processnow takesOption<&Indexing>#[non_exhaustive]; config enums gainPartialEq/Eq/Hash; fluent setters are#[must_use]