Corrected KPWr credits; link the training dataset

ATTRIBUTION.md

@@ -2,13 +2,15 @@
 
 v2 was fine-tuned from Matcha-TTS-PL (v1) on the fav2 dataset below. Both are credited.
 
+Note: the training data also included one KPWr 1.3 passage whose document record has no licence field (`kpwr-r20009`, popular-science subcorpus, source file badania.net.0.163.xml); KPWr 1.3 as a corpus is CC BY-SA 4.0. It is credited here and was left out of the published dataset.
+
 ## Training data of v2 (fav2 synthetic dataset)
 
-Per-clip records: `credits.jsonl` (6493 entries). Texts marked *adapted* were changed (numbers and abbreviations written out, a cut sentence completed, first-person forms made masculine, length trimmed).
+Per-clip records: `credits.jsonl` (6492 entries). Texts marked *adapted* were changed (numbers and abbreviations written out, a cut sentence completed, first-person forms made masculine, length trimmed).
 
 ## written for this dataset with Claude (Anthropic)
 
-3516 texts · licence: dataset licence
+3516 texts · licence: CC BY-SA 4.0 (original text)
 
 Original text written for this dataset (no third-party source).
 
@@ -375,7 +377,7 @@ Transcripts of YouTube videos published by their authors under CC BY 3.0, via th
 
 ## written for this dataset (tts-pl, voxcpm/build_texts.py)
 
-1246 texts · licence: dataset licence
+1246 texts · licence: CC BY-SA 4.0 (original text)
 
 Original text written for this dataset (no third-party source).
 
@@ -598,7 +600,7 @@ Original text written for this dataset (no third-party source).
 
 ## KPWr 1.3 (CLARIN-PL, hdl 11321/722)
 
-77 texts · licence: CC BY, CC BY 2.0 PL, CC BY 2.5 ES, CC BY 2.5 PL, CC BY 3.0 PL, CC BY-SA 2.5 PL, CC BY-SA 3.0, CC BY-SA 3.0 PL, CC Copyright-Only Dedication* (based on United States law) or Public Domain Certification (retired), genre = notatka prasowa
+76 texts · licence: CC BY, CC BY 2.0 PL, CC BY 2.5 ES, CC BY 2.5 PL, CC BY 3.0 PL, CC BY-SA 2.5 PL, CC BY-SA 3.0, CC BY-SA 3.0 PL, CC Copyright-Only Dedication* (based on United States law) or Public Domain Certification (retired)
 
 - `kpwr-r20061` „10 000 plików w kieszeni” · author: Alek Tarkowski · 2008-11-20 · https://kultura20.blog.polityka.pl/2008/11/20/10-000-plikow-w-kieszeni/ · doc 102375 · licence CC BY 2.5 PL
 - `kpwr-0025` „Andyjskie lody z miedzianej misy” · author: Ewa Kulak · 2012-09-07 · http://www.kolumbijsko.com/andyjskie-lody-z-miedzianej-misy/ · doc 103887 · licence CC BY 2.5 ES
@@ -654,7 +656,7 @@ Original text written for this dataset (no third-party source).
 - `kpwr-r20064` „Uczelniana Platforma e-Learningowa dla osób słabowidzących” · author: Michał (MO) · 2008-09-05 · https://www.cel.agh.edu.pl/uczelniana-platforma-e-learningowa-dla-osob-slabowidzacych/ · doc 101539 · licence CC BY 3.0 PL · adapted
 - `kpwr-r20004` „W Gdańsku o kulturze 2.0 – część II” · author: Alek Tarkowski · 2007-07-24 · https://kultura20.blog.polityka.pl/2007/07/24/w-gdansku-o-kulturze-20-czesc-ii/ · doc 102446 · licence CC BY 2.5 PL · adapted
 - `kpwr-r20036` „Web 2.0 – czy można ufać pośrednikom?” · author: Alek Tarkowski · 2010-12-27 · https://kultura20.blog.polityka.pl/2010/12/27/web-20-czy-juz-pora-zaczac-sie-bac/ · doc 103938 · licence CC BY 2.5 PL
-- `kpwr-r20052` „Wielkanocne ekspresje” · author: tokenization = wcrft:nkjp_s2.ini · 2012-11-26 · https://ozorkow.wimbp.lodz.pl/index.php/aktualnoci/212-wielkanocne-ekspresje.html · doc 103748 · licence CC BY-SA 2.5 PL · adapted
+- `kpwr-r20052` „Wielkanocne ekspresje” · 2012-11-26 · https://ozorkow.wimbp.lodz.pl/index.php/aktualnoci/212-wielkanocne-ekspresje.html · doc 103748 · licence CC BY-SA 2.5 PL · adapted
 - `kpwr-r20037` „Winda tymczasowa” · author: Michał Kowalski · 2007-01-20 · http://www.ozorkow.net/2007/01/20/winda-tymczasowa/ · doc 101872 · licence CC BY-SA 2.5 PL · adapted
 - `kpwr-0028` „Więcej zabawy, mniej gratów!” · author: Michał Parzuchowski · 2006-01-25 · http://badania.net/wiecej-zabawy-mniej-gratow/ · doc 102133 · licence CC BY-SA 3.0 PL · adapted
 - `kpwr-r20008` „Wykluczenie wykluczonych” · author: Rafał Próchniak · 2008-10-21 · http://web.archive.org/web/20111028094130/http://burlap.pl/wycinki/wykluczanie-wykluczonych · doc 101765 · licence CC BY-SA 2.5 PL · adapted
@@ -662,13 +664,12 @@ Original text written for this dataset (no third-party source).
 - `kpwr-r20047` „XXI wiek – Nowe Średniowiecze?” · author: Mirosław Filiciak · 2006-09-29 · https://kultura20.blog.polityka.pl/2006/09/29/xxi-wiek-%E2%80%93-nowe-sredniowiecze/ · doc 102073 · licence CC BY 2.5 PL
 - `kpwr-0015` „Xmods pierwszej generacji” · author: Redakcja serwisu www.stormdust.net · 2011-10-08 · http://www.stormdust.net/rc/pl/xmods/overview.html · doc 102391 · licence CC BY 2.0 PL · adapted
 - `kpwr-r20050` „Z kim walczy nowa kultura?” · author: Alek Tarkowski · 2007-01-08 · https://kultura20.blog.polityka.pl/2007/01/08/z-kim-walczy-nowa-kultura/ · doc 101807 · licence CC BY 2.5 PL
-- `kpwr-r20009` „badania.net.0.163.xml” · author: tokenization = wcrft:nkjp_s2.ini · 2011-10-08 · badania.net.0.163.xml · doc 102131 · licence genre = notatka prasowa
 - `kpwr-0050` „blog.przemekbicki.pl.0.3.xml” · author: Przemysław Bicki · 2011-10-08 · https://przemekbicki.pl · doc 101953 · licence CC BY
-- `kpwr-r20017` „burlap.pl.0.15.xml” · author: tokenization = wcrft:nkjp_s2.ini · 2007-07-16 · burlap.pl.0.15.xml · doc 102249 · licence CC BY-SA 2.5 PL · adapted
+- `kpwr-r20017` „burlap.pl.0.15.xml” · 2007-07-16 · burlap.pl.0.15.xml · doc 102249 · licence CC BY-SA 2.5 PL · adapted
 - `kpwr-0043` „cała moda węgierska” · author: Jeż Węgierski · 2013-02-22 · http://jezwegierski.blox.pl/html · doc 103916 · licence CC BY 3.0 PL · adapted
-- `kpwr-r20072` „fundacjanowemedia.org.0.24.xml” · author: tokenization = wcrft:nkjp_s2.ini · 2010-06-10 · fundacjanowemedia.org.0.24.xml · doc 102404 · licence CC BY-SA 3.0 PL
+- `kpwr-r20072` „fundacjanowemedia.org.0.24.xml” · 2010-06-10 · fundacjanowemedia.org.0.24.xml · doc 102404 · licence CC BY-SA 3.0 PL
 - `kpwr-r20048` „fundacjanowemedia.org.0.40.xml” · author: Redakcja serwisu fundacjanowemedia.org · 2005-05-31 · http://www.platerowka-szkola.edu.pl/public_html_old/Podstrony/gazeta.html · doc 102030 · licence CC BY-SA 3.0 PL
-- `kpwr-r20067` „fundacjanowemedia.org.0.42.xml” · author: tokenization = wcrft:nkjp_s2.ini · 2010-12-13 · http://redakcja.mam.media.pl/qmam-mam/10/6/4/ · doc 102361 · licence CC BY-SA 3.0 PL
+- `kpwr-r20067` „fundacjanowemedia.org.0.42.xml” · 2010-12-13 · http://redakcja.mam.media.pl/qmam-mam/10/6/4/ · doc 102361 · licence CC BY-SA 3.0 PL
 - `kpwr-0008` „iSummit” · author: Alek Tarkowski · 2007-06-12 · https://kultura20.blog.polityka.pl/2007/06/12/isummit/ · doc 102421 · licence CC BY 2.5 PL · adapted
 - `kpwr-r20005` „koleżanka korespondencyjna ze Związku Radzieckiego” · author: Jeż Węgierski · 2012-02-08 · http://jezwegierski.blox.pl/2012/02/kolezanka-korespondencyjna-ze-Zwiazku-Radzieckiego.html · doc 103928 · licence CC BY 3.0 PL · adapted
 - `kpwr-0023` „sos spaghetti i różnorodność społeczeństwa” · author: Alek Tarkowski · 2006-09-27 · https://kultura20.blog.polityka.pl/2006/12/27/sos-spaghetti-i-roznorodnosc-spoleczenstwa/ · doc 102419 · licence CC BY 2.5 PL · adapted

README.md

@@ -24,7 +24,8 @@ a 2B-parameter TTS, onto a single synthetic voice. What changes compared with v1
 - **A HiFi-GAN fine-tuned on this voice**, with real ground truth (the teacher's audio) — the v1 vocoder was tuned for the
   audiobook readers and blurs this voice.
 
-Try it in the browser: [spaces/machinekind/matcha-tts-pl-v2](https://huggingface.co/spaces/machinekind/matcha-tts-pl-v2).
+Try it in the browser: [spaces/machinekind/matcha-tts-pl-v2](https://huggingface.co/spaces/machinekind/matcha-tts-pl-v2) ·
+training data: [datasets/machinekind/polish-synthetic-speech-voxcpm2](https://huggingface.co/datasets/machinekind/polish-synthetic-speech-voxcpm2).
 
 ## Quality (12 held-out sentences, Mac CPU, 4 ODE steps, temperature 0.8)
 
@@ -76,7 +77,7 @@ yourself. The browser port produces the training phonemes character for characte
 
 - Base: Matcha-TTS-PL v1 (target model), new speaker row initialised from the closest real reader; same mel statistics,
   style tokens and phoneme table (all en-us phones were already in the symbol table, now trained).
-- Data: 6 493 clips, 13.1 h, generated by VoxCPM2 with the frozen fav2 voice (anchor + continuation, 10 diffusion
+- Data: [machinekind/polish-synthetic-speech-voxcpm2](https://huggingface.co/datasets/machinekind/polish-synthetic-speech-voxcpm2) — 6 493 clips, 13.1 h, generated by VoxCPM2 with the frozen fav2 voice (anchor + continuation, 10 diffusion
   steps), every clip gated by Whisper (text match), UTMOS ≥ 3.2 and speaker similarity ≥ 0.80, up to 3 takes per
   sentence. Source texts reviewed and minimally completed by Claude (cut sentences finished, numbers written as words).
 - Matcha fine-tune: 10 k steps, batch 64, lr 5e-5, bf16, RTX 6000 Ada (≈ 1.5 h); quality plateaued from 5 k steps.