Changelog

Added NumPy-only spectral analysis, pitch tracking, reference metrics, and STOI/ESTOI APIs

Changed source distributions to exclude project-local ledger state

Documented analysis APIs, metric formulas, validation rules, and quality-evaluation boundaries

Improved regression and quality coverage for spectral, pitch, metric, intelligibility, and packaging behavior

Added reusable selective loudness analysis, gain-adjusted metrics, and memory-bounded 4x true-peak measurement

Added time-varying speech rate and pitch envelope APIs with exact output-frame prediction

Documented envelope controls, timing semantics, backend behavior, and perceptual limitations

Improved regression coverage and local evaluation tools for envelope timing, trajectories, and transitions

Added BS.1770-style integrated loudness and dBFS/dBTP peak measurement APIs

Documented loudness, peak, RMS, normalization, and downstream import boundaries

Improved loudness and true-peak conformance coverage

Added deterministic NumPy-only smooth waveform cut-point selection within caller-provided sample intervals

Documented smooth cut-point semantics, complexity, examples, and downstream policy boundaries

Improved boundary quality coverage across validation, downstream contracts, packaging, axes, and dtypes

Added NumPy-only silence generation and explicit-axis mono downmixing APIs for TTSForge compatibility

Added deterministic silence generation and explicit channel-axis arithmetic downmixing for portable TTS

Phase-vocoder time stretching with configurable window size and overlap

Windowed-sinc resampler with rounded output sizing and an optional librosa-compatible ceiling mode

Pitch shifting via resample-then-stretch that preserves exact input length

Normalized-energy and relative-dB voice activity detection algorithms

RMS amplitude normalization and peak amplitude utilities

Silence trimming and audio splitting with configurable frame parameters

Audio framing, short-time energy, zero-crossing rate, and spectral flux utilities

Typed AudioSignalError exceptions for invalid arrays and parameters

NumPy-only dependency with no native extensions required

NumPy-only WSOLA speech backend with explicit backend selection

Experimental NumPy-only ESOLA speech backend with exact-length synthesis, epoch alignment, and explicit moderate-rate validation

Experimental NumPy-only direct TD-PSOLA speech pitch/prosody backend with bounded pitch tracking, voiced-grain synthesis, and WSOLA unvoiced fallback

Single-pass pitch/rate compositor planner and deterministic quality harness

README overview plus detailed API, getting-started, advanced, and examples documentation

Backend trade-offs, exact-length semantics, rolloff propagation, and native formant limitations

ESOLA speech-only scope, rate limits, independent-lane behavior, and listening-evaluation gate

TD-PSOLA experimental scope, conservative limits, fallback behavior, and real-speech listening-evaluation gate

Full test suite with quality, compatibility, and packaging tests

Deterministic source-filter-style speech fixtures and local WAV comparison metrics