Skip to content

sadda

Sadda (Pali सद्द — "sound, voice") is an open-source toolkit for phonetics and speech-science research. It targets the same audience as Praat but with a modern Python API, NumPy- and Polars-native data types, and a corpus-first project model.

The sadda desktop app showing a speech recording as a waveform, a spectrogram, and stacked f0, formant, and intensity measure tracks, alongside a bundle-list sidebar and a reference-distribution panel.

These docs always reflect the latest commit on main. For per-release notes, see the CHANGELOG and GitHub releases.

Install

pip install sadda           # core: corpus, DSP, clinical, refdist, recipes
pip install "sadda[ml]"     # also installs onnxruntime — VAD + embeddings

Pre-built wheels are available for Linux x86_64, macOS arm64, and Windows x86_64 on Python 3.10–3.13. Other platforms install from sdist and need a Rust toolchain locally.

The [ml] extra adds ONNX Runtime (~25 MB). The base install is lean for users who don't need VAD / embeddings; ML calls raise a clean "ONNX Runtime not available" error if the extra isn't installed.

What's here

Stability tiers

Per the 2026-05-18 Python API surface entry (in DEVLOG.md):

Tier Modules Commitment
Stable sadda.corpus, sadda.dsp, top-level project loaders Won't break across minor versions
Stable (clinical) sadda.clinical Same commitment as Stable; the separate tier flags that these measures are research-use only, not for clinical diagnosis
Provisional sadda.live, sadda.recipe, sadda.refdist, sadda.ml May break in minor versions after a deprecation cycle
Experimental sadda.experimental.* (none yet) May break any release

Non-stable APIs emit a one-time ProvisionalAPIWarning / ExperimentalAPIWarning on first use (suppressible via the standard warnings machinery).

Source

Everything lives at github.com/sadda-speech/sadda. The DEVLOG.md in the repo is the running design-decision log; the docs here are a curated subset of what's most useful when you're actually using the library.

License

Dual-licensed under Apache 2.0 OR MIT.