LoginSign Up
← Back to all sparks
E

emuR

INFRA · APIS
Velocity0.0

Main Package of the EMU Speech Database Management System

github.com/ips-lmu/emur ↗

The R half of the EMU speech database system, fixing what was quietly broken.

speech-sciencephoneticsannotationr-packagecorpus-toolsdatabase
◆Current state
emuR is the R interface to the EMU Speech Database Management System — loading annotated speech corpora, running hierarchical queries over annotation levels, extracting signal track data, and serving corpora to the EMU-webApp for browser-based annotation. It is at 2.6.0 on a slow cadence of roughly one release a year. Recent work has centred on the CRUD operations for annotation items and on widening what serve() can hand the web application.
◆Where it's heading
The releases read as a package being brought up to the standard its own API implied. delete_itemsInLevel() shipped in 2.1.1 as a first version, was described in 2.5.0 as heavily flawed and now usable, and the create/update/delete family is still called ongoing work. Alongside that, the query engine was rewritten onto CTEs and the signal-processing layer is being opened past the bundled wrassp, starting with Matlab. Speed work recurs — SQLite transactions, prepared statements, on-the-fly caching — consistent with corpora outgrowing the original design.
◆Prediction
Two threads are explicitly unfinished: the CRUD documentation and behaviour, described as ongoing, and the add_signalVia family, described as a draft starting with Matlab. Expect the next release to advance one of them rather than open new ground.

◆Recent moves

  1. 8mo ago

    emuR 2.6.0

    serve() accepts signal tracks stored as Rda and converts them to SSFF on the fly, removing a manual conversion step before corpora can be opened in the web application. Continues the pattern of widening what the browser-facing half will accept.

    View source ↗
  2. 1y ago

    emuR 2.5.1

    Drafts an add_signalVia family so signal processing can come from outside wrassp, beginning with Matlab. Explicitly a draft, but it is the first move toward treating the bundled signal-processing library as one option rather than the only one.

    View source ↗
  3. 1y ago

    emuR 2.5.0

    delete_itemsInLevel() is described as having been heavily flawed and now usable, and create_itemsInLevel() accepts more calling patterns. A core annotation-editing operation moving from nominally present to actually dependable, which is the thread running through this package's recent history.

    View source ↗
  4. 3y ago

    emuR 2.4.0

    Caches on-the-fly computations in get_trackData(), fixes a level attribute mixup in seglist generation, and works around a wrassp call that could crash the R session. Also the tidyverse 2.0.0 adaptation pass that kept CRAN checks green.

    View source ↗
  5. 5y ago

    emuR 2.3.0

    A long list of guard rails and ergonomics: directory checks, session and bundle pattern filters for listing, clearer warnings about heterogeneous sampling rates, cli-based summary output, and a query speedup from avoiding full table scans. Corpus-scale usability work rather than new capability.

    View source ↗
  6. 5y ago

    emuR 2.2.0

    The largest structural change in this window: hierarchical queries and requeries are rewritten on common table expressions and the EQL engine is cleaned up. Bundle saving and corpus serving get SQLite transaction and prepared-statement speedups, and spectrograms can be rendered as rasters using the web app's own drawing code.

    View source ↗