# Sources and rights

## Current OWL / ElevenLabs presentation

Words and lexical values are supplied by the project owner in stim-banks/owl-bank.csv. The CSV SHA256 and exact lexical values are retained in owl/manifest.json and run exports. No external source citation or license was provided for these measures; none is inferred. Missing categories remain missing.

Current speech files under owl/male and owl/female were generated with the supplied ElevenLabs voices using eleven_multilingual_v2. Model, voice IDs, settings, hashes, durations and processing are recorded in manifests. These files are not the CC0 eSpeak outputs described below; no open-content license is asserted for ElevenLabs output. The app serves static files and does not send participant data to ElevenLabs. Pronunciation, intelligibility and physical playback timing require human/device review.

## Legacy wordfreq / eSpeak bank

Frequency subset and curated bank: CC BY-SA 4.0, https://creativecommons.org/licenses/by-sa/4.0/ . Derived from Robyn Speer et al., wordfreq 3.0.2, pinned commit 372f6dbb3bd2ad4b675a3b18b94d44d0dd5fea8b. Numeric subset selected and annotated with categories; not a claim of validated equivalent lists. The complete source attribution/citations are retained in wordfreq-README.md, and code license in wordfreq-MIT-LICENSE.txt.

Credit Google Books Ngrams (http://books.google.com/ngrams), Leeds Internet Corpus, Wikipedia, ParaCrawl, OPUS/OpenSubtitles, and freely available SUBTLEX data by Marc Brysbaert and colleagues (http://crr.ugent.be/programs-data/subtitle-frequencies). See the retained README for complete constituent corpus citations and licensing conditions. Source and subset hashes are in the exported bank.

Audio: original synthetic WAV output generated locally with eSpeak NG 1.51 (https://github.com/espeak-ng/espeak-ng), en-us voice, 145 words/minute. Program GPLv3+; output is not a derivative of the program (https://www.gnu.org/licenses/gpl-faq.html#GPLOutput). No proprietary clinical list or third-party recording used. These generated recordings are made available under CC0 1.0 (https://creativecommons.org/publicdomain/zero/1.0/). Fixed audio content, not browser TTS. Intelligibility and physical playback latency remain unvalidated.

Roboto font: Apache 2.0, see ../ROBOTO-LICENSE.txt. CanvasKit: existing suite's m2c2kit/Skia WASM dependency; BSD-style Skia license https://skia.googlesource.com/skia/+/main/LICENSE .
