# Changelog

## Score

- CAI 37 → 35 (-2.4)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.4) — scores are not directly comparable.

## Lenses

- Code Health 66 → 61 (-5.8)
- Architecture 92 → 93 (+0.6)
- Maturity 57 → 58 (+0.4)
- Readiness 35 → 34 (-1.3)
- Security 28 → 22 (-5.6)
- Accessibility 44 → 49 (+5.0)

## Resolved (19)

- Change coupling: train_hifigan.py ↔ train_multiband_melgan.py (recipes/ljspeech/hifigan/train_hifigan.py)
- Change coupling: train_multiband_melgan.py ↔ train.py (recipes/ljspeech/multiband_melgan/train_multiband_melgan.py)
- Change-coupling hub: train_glowtts.py → train_aligntts.py, train_hifigan.py, train_multiband_melgan.py (recipes/ljspeech/glow_tts/train_glowtts.py)
- Change-coupling hub: train_speedy_speech.py → train_aligntts.py, train_fast_pitch.py, train_glowtts.py, train_vits.py, train_vits_tts.py, train_vits.py (recipes/ljspeech/speedy_speech/train_speedy_speech.py)
- Change-coupling hub: train_vits.py → train_fast_pitch.py, train_vits.py, train_vits_tts.py (recipes/vctk/vits/train_vits.py)
- Duplicated block (11 lines × 2) (TTS/tts/models/neuralhmm_tts.py)
- Duplicated block (5 lines × 2) (TTS/encoder/models/resnet.py)
- Duplicated block (5 lines × 3) (TTS/tts/datasets/dataset.py)
- Duplicated block (6 lines × 2) (TTS/tts/datasets/formatters.py)
- Duplicated block (7 lines × 2) (TTS/tts/layers/xtts/hifigan_decoder.py)
- Duplicated block (7 lines × 2) (TTS/tts/models/delightful_tts.py)
- Duplicated block (7 lines × 2) (TTS/tts/models/forward_tts.py)
- Duplicated block (8 lines × 2) (TTS/encoder/models/resnet.py)
- Inconsistent parameter ordering and naming for file output. `tts_to_file` adds `pipe_out` and `file_path` but keeps the same positional argument order as `tts`, which is confusing. Furthermore, `tts` returns audio (implied), while `tts_to_file` writes to disk. The presence of `pipe_out` in `tts_to_file` suggests a hybrid behavior that is not present in the base `tts` method, creating a split in how output is handled.
- Near-duplicate member pair (43 shared lines) (TTS/encoder/models/resnet.py)
- No ADRs found
- Redundant `_to_file` variant for TTS with Voice Conversion. Same inconsistency as above.
- Redundant loading methods with overlapping signatures. `load_vc_model_by_name` and `load_tts_model_by_name` are specific variants of `load_model_by_name`. This forces users to guess which method to call based on the model type, rather than using a single unified entry point that handles dispatch internally.
- Same pattern as TTS: redundant `_to_file` variants. `voice_conversion` and `voice_conversion_to_file` duplicate intent. The base method likely returns audio data, while the `_to_file` variant writes it. This pattern is repeated for `tts` and `tts_with_vc`.

## New (121)

- Change coupling: train_aligntts.py ↔ train_glowtts.py (recipes/ljspeech/align_tts/train_aligntts.py)
- Change-coupling hub: train_multiband_melgan.py → train_glowtts.py, train_hifigan.py, train.py (recipes/ljspeech/multiband_melgan/train_multiband_melgan.py)
- Change-coupling hub: train_speedy_speech.py → train_fast_pitch.py, train_vits.py, train_vits.py (recipes/ljspeech/speedy_speech/train_speedy_speech.py)
- Edited copies of one member (3 members, 11+ corresponding lines) (TTS/tts/datasets/formatters.py)
- Edited copies of one member (3 members, 11+ corresponding lines) (TTS/tts/layers/vits/discriminator.py)
- Edited copies of one member (3 members, 13+ corresponding lines) (TTS/tts/models/overflow.py)
- Edited copies of one member (3 members, 17+ corresponding lines) (TTS/tts/datasets/formatters.py)
- Edited copies of one member (3 members, 17+ corresponding lines) (TTS/tts/datasets/formatters.py)
- Edited copies of one member (3 members, 19+ corresponding lines) (TTS/tts/layers/xtts/hifigan_decoder.py)
- Edited copies of one member (3 members, 3+ corresponding lines) (TTS/tts/layers/vits/networks.py)
- Edited copies of one member (3 members, 31+ corresponding lines) (TTS/tts/layers/xtts/hifigan_decoder.py)
- Edited copies of one member (3 members, 35+ corresponding lines) (TTS/tts/layers/delightful_tts/conv_layers.py)
- Edited copies of one member (3 members, 7+ corresponding lines) (TTS/tts/layers/delightful_tts/acoustic_model.py)
- Edited copies of one member (3 members, 9+ corresponding lines) (TTS/tts/models/delightful_tts.py)
- Edited copies of one member (4 members, 3+ corresponding lines) (TTS/vocoder/models/fullband_melgan_generator.py)
- Edited copies of one member (4 members, 5+ corresponding lines) (TTS/tts/models/delightful_tts.py)
- Edited copies of one member (5 members, 3+ corresponding lines) (TTS/tts/models/align_tts.py)
- Edited copies of one member (7 members, 3+ corresponding lines) (TTS/tts/models/align_tts.py)
- Edited copy of a member (10 corresponding lines) (TTS/tts/models/vits.py)
- Edited copy of a member (11 corresponding lines) (TTS/tts/layers/bark/model.py)
- …and 101 more

## Architecture

- Unchanged — 0 containers · 1 contexts · 0 edges
