Most v1.6.1 code remains source-compatible. The changes below affect explicit types, latency assumptions or thread ownership and should be reviewed during an upgrade.

Analysis-window latency

Default analysis windows now preserve approximately the same time span across sample rates instead of preserving a fixed sample count. This keeps frequency register and spectral resolution consistent, but increases the resolved window at higher rates.

Callers that relied on default-window latency at sample rates at or above 88.2 kHz may pass windowSize=4096 explicitly and accept the documented reduced register. Use getWindowSize() for latency accounting rather than assuming a fixed default. Components that name the parameter fftSize expose the equivalent resolved-size getter documented by that component.

Explicit positive window and hop values remain available when a fixed count is part of an application's contract. Above the automatic window's documented sample-rate ceiling, choose an explicit size if the application needs a lower frequency floor.

Biquad coefficients and publication ownership

BiquadCoeffs<T> is now the non-template BiquadCoeffs; its fields and the biquad recursion use double precision regardless of the buffer sample type. For example:

// v1.6.1
auto coefficients = dspark::BiquadCoeffs<float>::makeLowPass(48000.0, 1200.0);
 
// v1.7.0
auto coefficients = dspark::BiquadCoeffs::makeLowPass(48000.0, 1200.0);

The former BiquadCoeffs::makeDcBlocker() convenience factory is no longer part of the coefficient API. Use DCBlocker<T> for DC removal, or BiquadCoeffs::makeHighPass() when a general biquad high-pass is specifically required.

Choose the coefficient setter by origin:

setCoeffs() publishes a set from a control thread to the processing thread. Adoption on the audio thread is bounded; if publication overlaps every read attempt, the processor keeps its previous complete set, re-arms the pending update and adopts it on a later call.
setCoeffsNow() directly updates audio-thread-private state. Use it only for coefficients computed by the stream owner, or in single-threaded/offline processing. It is not a replacement for concurrent control-thread publication.

Designate one master for each embedded Biquad: either control-published or stream-owner-modulated. Mixing both origins on one instance makes the last set adopted by the audio thread win and obscures ownership.

<tt>AudioBufferView</tt> type queries

The converting and pointer-array constructors now constrain the pointer conversion instead of failing with a check inside the constructor body. A mutable view still converts to a const view; the reverse now correctly fails during overload resolution and in type traits:

static_assert(std::is_convertible_v<
    dspark::AudioBufferView<float>,
    dspark::AudioBufferView<const float>>);
 
static_assert(!std::is_convertible_v<
    dspark::AudioBufferView<const float>,
    dspark::AudioBufferView<float>>);

No previously valid evaluated conversion changes meaning. Code that depended on the old false-positive result in std::is_convertible_v, std::is_constructible_v, decltype, SFINAE or a requires expression should query the legal mutable-to-const direction instead.

AudioBuffer<T, MaxChannels>::toView() now preserves MaxChannels in the returned view type. Prefer auto for the result, or spell the matching AudioBufferView<T, MaxChannels> type when an exact type is required.

<tt>AudioProcessor</tt> means in-place

AudioProcessor now accepts only processors whose processBlock contract is unambiguously an in-place mutable-buffer operation. Read-only analysers such as BeatTracker, LoudnessMeter, OnsetDetector and PitchFollower no longer satisfy it; run them beside a ProcessorChain on a const view of the same buffer.

A type exposing both mutable and const processBlock overloads is treated as read-only by the concept. If the type is intended to be an insert, give its analysis operation a distinct name.