Renamed from Matcha-TTS-PL-v2 to Matcha-TTS-PL; base model is Matcha-TTS-PL-Pretrain

ATTRIBUTION.md

@@ -1,8 +1,8 @@
-# Attribution — Matcha-TTS-PL v2
+# Attribution — Matcha-TTS-PL
 
-Every voice of v2 was fine-tuned from Matcha-TTS-PL (v1) on its own recordings of the same texts (dataset below, one subset per voice). Both are credited.
+Every voice was fine-tuned from Matcha-TTS-PL-Pretrain (formerly Matcha-TTS-PL) on its own recordings of the same texts (dataset below, one subset per voice). Both are credited.
 
-Note: the training data of male_voxcpm_1 (the first v2 voice) 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 is not in the published dataset or the later voices.
+Note: the training data of male_voxcpm_1 (the first voice) 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 is not in the published dataset or the later voices.
 
 ## Training data (machinekind/polish-synthetic-speech-voxcpm2, all voices)
 
@@ -936,7 +936,7 @@ Original text written for this dataset (no third-party source).
 
 ---
 
-## Base model (Matcha-TTS-PL v1): Atrybucja danych treningowych — Matcha-TTS-PL (v3)
+## Base model (Matcha-TTS-PL-Pretrain): Atrybucja danych treningowych — Matcha-TTS-PL (v3)
 
 Wagi modelu: **CC BY-SA 4.0** (`LICENSE-MODEL`). Poniżej źródła nagrań, ich licencje i wprowadzone zmiany.
 Głos modelu jest syntezą (interpolacja embeddingów wielu mówców) i nie odtwarza głosu żadnej konkretnej osoby.

README.md

@@ -3,13 +3,15 @@ language: pl
 license: cc-by-sa-4.0
 library_name: matcha-tts
 pipeline_tag: text-to-speech
-base_model: machinekind/Matcha-TTS-PL
+base_model: machinekind/Matcha-TTS-PL-Pretrain
 tags: [tts, polish, matcha-tts, flow-matching, hifigan, distillation, voxcpm2, code-switching, robot, onnx, multi-voice]
 ---
 
-# Matcha-TTS-PL v2 — four synthetic voices, modern Polish, English words
+# Matcha-TTS-PL — four synthetic voices, modern Polish, English words
 
-Second version of [Matcha-TTS-PL](https://huggingface.co/machinekind/Matcha-TTS-PL): the same small, fast,
+*Formerly published as `machinekind/Matcha-TTS-PL-v2` (old links redirect here).*
+
+Fine-tuned from [Matcha-TTS-PL-Pretrain](https://huggingface.co/machinekind/Matcha-TTS-PL-Pretrain) (pretraining on Polish audiobooks: Wolne Lektury + AZON): a small, fast,
 non-autoregressive Matcha-TTS (20.9 M parameters, real-time on a small GPU or in a browser) **distilled from VoxCPM2**,
 a 2B-parameter TTS, onto synthetic voices. **One model per voice** (each with its own fine-tuned HiFi-GAN vocoder), all in
 this repo:
@@ -33,7 +35,7 @@ description with VoxCPM2 (OpenBMB, Apache-2.0) and frozen as prompt caches; not
   English words are detected automatically and phonemized by espeak-ng en-us, the rest by espeak-ng pl (`english/`).
 - **A HiFi-GAN fine-tuned on each voice**, with real ground truth (the teacher's audio).
 
-Try it in the browser (voice selector): [spaces/machinekind/matcha-tts-pl-v2](https://huggingface.co/spaces/machinekind/matcha-tts-pl-v2) ·
+Try it in the browser (voice selector): [spaces/machinekind/matcha-tts-pl](https://huggingface.co/spaces/machinekind/matcha-tts-pl) ·
 training data: [datasets/machinekind/polish-synthetic-speech-voxcpm2](https://huggingface.co/datasets/machinekind/polish-synthetic-speech-voxcpm2).
 Defaults: **temperature 1.0**, the voice's **tempo** above (`length_scale`; at 1.0 the models speak 10–19 % slower than the teacher), style token 8 = neutral.
 
@@ -83,7 +85,7 @@ VoxCPM2's question intonation (wh-questions fall, yes/no questions rise); a spac
 
 ## Training
 
-- Base: Matcha-TTS-PL v1 (target model), new speaker row initialised from the closest real reader; same mel statistics,
+- Base: [Matcha-TTS-PL-Pretrain](https://huggingface.co/machinekind/Matcha-TTS-PL-Pretrain) (its 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: [machinekind/polish-synthetic-speech-voxcpm2](https://huggingface.co/datasets/machinekind/polish-synthetic-speech-voxcpm2) — about 8 000 clips (≈ 15 h) per voice, generated by VoxCPM2 with each frozen voice (anchor + continuation, 10 diffusion
   steps), every clip gated by Whisper (text match), UTMOS ≥ 3.2, speaker similarity ≥ 0.80 and pitch, up to 3 takes per

samples/README.md

@@ -1,6 +1,6 @@
-Held-out test sentences (not in the training set). v2 at tempo 0.8, temperature 0.8, 4 ODE steps (the Space default temperature is now 1.0); teacher = VoxCPM2 with the frozen male_voxcpm_1 voice.
+Held-out test sentences (not in the training set). male_voxcpm_1 at tempo 0.8, temperature 0.8, 4 ODE steps (the Space default temperature is now 1.0); teacher = VoxCPM2 with the frozen male_voxcpm_1 voice.
 
-| # | text | v2 | VoxCPM2 teacher |
+| # | text | model | VoxCPM2 teacher |
 |---|---|---|---|
 | e01 | Dzień dobry! W czym mogę dziś pomóc? | [v2](samples/e01_v2.wav) | [teacher](samples/e01_teacher.wav) |
 | e02 | Proszę zamknąć okno, robi się zimno. | [v2](samples/e02_v2.wav) | [teacher](samples/e02_teacher.wav) |