Licenses & data sources

Word levels

"How hard is this?" gives every word a level from A1 to C2. Two lists supply the levels, and they are kept apart, because they come under different terms.

The credits, as the apps show them

Download the modified C1–C2 list

• TSV octanove_c1_c2.tsv2,119 rows, 1,998 words · word, part of speech, level · 41 KB
• TXT LICENSE-octanove-CC-BY-SA-4.0.txtwhat we changed, and the full license · 22 KB

This file is a modified version of the Octanove Vocabulary Profile C1/C2, version 1.0, by Octanove Labs, and is licensed under the Creative Commons Attribution-ShareAlike 4.0 International license (CC BY-SA 4.0), the same license as the original. You may reuse it under that same license: credit Octanove Labs and NeuroPhonetics LLC, say whether you changed it, and if you share an adapted version, share it under CC BY-SA 4.0 too. The license gives no warranties.

What we changed — the levels themselves were not changed:

• surrounding spaces trimmed (2 headwords);
• headwords with alternatives separated by "/" split into one row per spelling, with the same part of speech and level (46 headwords);
• the notes column (sense glosses) dropped, and rows that were then the same word with the same part of speech merged, keeping the lowest level (63 rows);
• two part-of-speech labels corrected: "remonstrate" vern → verb; "batter" (one who bats) empty → noun.

The original: github.com/openlanguageprofiles/olp-en-cefrj, file octanove-vocabulary-profile-c1c2-1.0.csv.

The CEFR-J Wordlist is not redistributed here

Levels A1–B2 come from The CEFR-J Wordlist Version 1.6, compiled by Yukio Tono, Tokyo University of Foreign Studies (Tono Laboratory, TUFS). Its terms allow research and commercial use with a proper acknowledgement of the source; they are not a Creative Commons license. So the apps carry the list with the credit above, and this site does not offer our copy for download. To get the list, go to its own page: cefr-j.org/download.html.

In our copy, alternatives separated by "/" became one row each (179 headwords), spellings were lower-cased (64), two duplicate rows were merged at the lowest level, and the columns CoreInventory 1, CoreInventory 2 and Threshold were dropped. Parts of speech and levels are unchanged.

Not endorsed. The levels are adapted by NeuroPhonetics LLC. Neither the CEFR-J project (Tono Laboratory, TUFS) nor Octanove Labs endorses, supervises or is otherwise associated with NeuroPhonetics LLC or its apps.

Pronunciation data

The broad IPA in every app that shows IPA starts from the CMU Pronouncing Dictionary (Carnegie Mellon University): each word is CMU's, written out in IPA by Phonema's rules, so every app gives the same answer.

The table the apps carry on the phone was built from the dictionary as packaged in the cmudict Python package, version 1.1.3 (135,166 entries). Its license ships inside each app that carries the table, and reads:

Copyright (C) 1993-2015 Carnegie Mellon University. All rights reserved.

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:

1. Redistributions of source code must retain the above copyright
   notice, this list of conditions and the following disclaimer.
   The contents of this file are deemed to be source code.

2. Redistributions in binary form must reproduce the above copyright
   notice, this list of conditions and the following disclaimer in
   the documentation and/or other materials provided with the
   distribution.

This work was supported in part by funding from the Defense Advanced
Research Projects Agency, the Office of Naval Research and the National
Science Foundation of the United States of America, and by member
companies of the Carnegie Mellon Sphinx Speech Consortium. We acknowledge
the contributions of many volunteers to the expansion and improvement of
this dictionary.

THIS SOFTWARE IS PROVIDED BY CARNEGIE MELLON UNIVERSITY ``AS IS'' AND
ANY EXPRESSED OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY
NOR ITS EMPLOYEES BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Model voices

Two kinds of voice: four recorded people, and a machine. Sentio says which is which wherever it tells a person from a machine.

• bdlman · recorded
• rmsman · recorded
• clbwoman · recorded
• sltwoman · recorded
• Neural2-Dsynthetic
• Neural2-Fsynthetic

The recorded voices — CMU ARCTIC

The model recordings a learner hears in Sentio, and the reference vowel chart a learner is judged against, are derived from the CMU ARCTIC speech databases (version 0.95), recorded at the Language Technologies Institute, Carnegie Mellon University, by John Kominek and Alan W Black. We use four General American voices: bdl and rms (men), clb and slt (women). The vowel-space figures on this site are measured from the same four voices.

These recordings are modified. No original ARCTIC file is distributed. Each clip is a short fragment cut out of a recorded sentence around one vowel and its neighboring consonant, trimmed at its edges, with its loudness leveled against the other clips. The reference chart is not audio at all but numbers measured from those recordings: the first two formants of each vowel, and their averages per vowel class.

The databases are distributed under these terms, from Appendix A of the technical report CMU ARCTIC databases for speech synthesis (CMU-LTI-03-177):

Carnegie Mellon University
Copyright (c) 2003
All Rights Reserved.

Permission to use, copy, modify, and license this software and its documentation for any
purpose, is hereby granted without fee, subject to the following conditions:

1. The code must retain the above copyright notice, this list of conditions and the following
   disclaimer.
2. Any modifications must be clearly marked as such.
3. Original authors' names are not deleted.

THE AUTHORS OF THIS WORK DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ALL
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR
ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS
OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS
ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.

The synthetic voices — Google Cloud Text-to-Speech

The remaining model clips were made for this project with Google Cloud Text-to-Speech, voices en-US-Neural2-D and en-US-Neural2-F. They are synthetic, and marked as such. The spectrograms and sound diagrams on this site are drawn from clips made with the same service.

Sentio also checks its own measurements against published vowel averages from Hillenbrand, Getty, Clark and Wheeler, Acoustic characteristics of American English vowels, Journal of the Acoustical Society of America 97(5), 1995. Only statistics derived from that paper are used, never its recordings, and never as the target a learner is judged against.

Tongue images — real-time MRI

The tongue shapes in Phonema ("In the mouth") and the moving mouth clips in Sentio come from the USC 75-Speaker Speech MRI Database: Y. Lim, A. Toutios et al., "A multispeaker dataset of raw and reconstructed speech production real-time MRI video and 3D volumetric images," Scientific Data 8, 187 (2021). The dataset: doi.org/10.6084/m9.figshare.13725546.

These images are modified. The tongue contours are traced by NeuroPhonetics from the published videos; Sentio's clips are silent, looping and resampled to 25 frames per second. Every image in the apps carries the line "MRI: USC 75-Speaker (Lim, Toutios et al. 2021), CC BY 4.0, contours traced by NeuroPhonetics".

Licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Dictionary meanings

The short meaning under a word comes from the Open English WordNet, 2025 edition, shortened to the commonest senses. It is looked up on the phone; the word never leaves it.

The Open English WordNet is licensed under Creative Commons Attribution 4.0 (CC BY 4.0) and is derived from Princeton WordNet under the WordNet License. To choose which words are common enough to carry, we also use Princeton WordNet 3.0's sense counts (from SemCor), under the WordNet License. Sources: en-word.net · wordnet.princeton.edu.

Open-source software

Each app lists the open-source software it is built with, and the license of each, in its own settings, under Open-source licenses.

This website uses two typefaces, Fraunces and Figtree, under the SIL Open Font License: OFL-Fraunces.txt · OFL-Figtree.txt.

A question about any of this: .