888 Commits
Author SHA1 Message Date
Deniz Şafak 1807822f58 Merge pull request #225 from denniswebb/publish/fix-alphanumeric-footnotes
fix(normalization): preserve alphanumeric terms during footnote cleanup
2026-11-01 20:25:07 +03:00
Deniz Şafak f39837dc3a Merge pull request #223 from denniswebb/publish/fix-m4b-cover-metadata
fix(exporters): preserve M4B cover artwork and metadata
2026-11-01 20:24:36 +03:00
Deniz Şafak 628f99d9b3 Merge pull request #222 from denniswebb/publish/fix-wizard-chapter-chunks
fix(webui): synchronize wizard chunks by source chapter
2026-11-01 20:24:09 +03:00
Deniz Şafak 42b430f077 Merge pull request #219 from denniswebb/publish/fix-kokoro-optional-speed
fix: avoid optional Kokoro speed formatting crash in WebUI
2026-11-01 20:23:27 +03:00
agent 4dd0bf0ff2 fix(normalization): preserve unmarked alphanumeric terms
Only remove explicit bracketed footnote markers. Plain letter-digit text cannot distinguish identifiers from flattened footnotes. Update the old suffix-removal expectation and cover the intentional superscript compatibility change.
2026-10-28 21:31:41 -05:00
agent 2b53256d43 test(normalization): reproduce alphanumeric footnote corruption 2026-10-28 21:30:11 -05:00
agentandClaude Opus 5.5 4a980856bd fix(webui): synchronize wizard chunks by source chapter
Synchronize selected chapter ordinals at update and finish while retaining prepared narration and speakers for unchanged source chapters at the same chunk level. Rebuild only newly selected or changed chapters, and propagate chapter voice overrides.

Add synthetic regression coverage for selection toggles, prepared fields, chunk levels, normalization calls and an opt-in EPUB check.

Fixes #220

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-28 21:21:43 -05:00
agentandClaude Opus 5.5 0eeeb53d11 fix(exporters): preserve M4B cover artwork and metadata
Map the cover after the optional FFmetadata input and omit use_metadata_tags only for cover-bearing exports. Preserve non-empty supplied text values as UTF-8 MP4 freeform tags alongside standard atoms using existing Mutagen.

Add command-line regression cases and real-FFmpeg round trips for artwork, chapters, audio, and custom metadata. The no-cover path is unchanged.

Fixes #221

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-28 21:21:41 -05:00
agentandClaude Opus 5.5 ec4e3c5735 fix: log optional Kokoro voice speeds safely
Use %s for three Optional[float] diagnostics so the WebUI Rich handler does not turn a valid speed=None into a conversion failure. Preserve request-speed fallback semantics and add focused regression coverage.

Fixes #218

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-28 21:17:50 -05:00
Deniz Şafak c2af9907f2 Merge pull request #216 from abaichuan1017/fix/atomic-pronunciation-writes
Keep pronunciation overrides intact when saving fails
2026-10-25 14:50:32 -07:00
Kiven 9cd295dbf4 fix: atomically replace pronunciation overrides 2026-10-26 00:15:39 +08:00
Deniz Şafak c0e9f14a43 Merge pull request #215 from MarkMad/perf/epub-open-209
perf(epub): speed up opening large EPUBs
2026-10-24 04:48:01 -07:00
MarkMadandClaude Sonnet 5.5 e2c451ba11 perf(epub): speed up opening large EPUBs
Opening a large EPUB spent most of its time in two linear scans repeated
once per navigation entry:

- _find_doc_key scanned every spine document for a basename match.
- spine_docs.index() / _docs_between looked up spine positions by list scan.

Build a basename lookup once per doc_order mapping and a
{document: first spine index} map, in both the PyQt parser (book_parser.py)
and the WebUI extractor (text_extractor.py). Exact and URL-decoded matches
are still tried first, and spine positions keep list.index() first-occurrence
semantics.

On a 41 MB EPUB with 7,520 chapters the PyQt parser drops from ~147 s to
~19 s and the WebUI extractor from ~30 s to ~19 s, with byte-identical
extracted text.

Partially addresses #209 (suggestions 1 and 2).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-10-24 15:27:17 +11:00
Deniz Şafak 654ec6d6fd Merge pull request #213 from chiragjangiir/fix/pasted-text-metadata-194
fix: preserve pasted text title in chapter metadata
2026-10-17 08:41:21 -07:00
Chirag Jangir 8a1d95be26 fix: preserve pasted text title in chapter metadata 2026-10-17 15:23:11 +05:30
Deniz Şafak 349fb8c233 Merge pull request #205 from MohammadHijjawi97/fix-output-layout-unique-base
fix(output): use collision-free merged path so repeat conversions don't overwrite
2026-10-15 07:36:00 -07:00
Deniz Şafak 8b3f7fcaa0 Merge pull request #211 from hari2m/fix/audio-sink-fail-fast
fix(audio_sink): fail fast when the merged-output ffmpeg encode fails
2026-10-15 07:33:25 -07:00
Deniz Şafak 9deda55296 Merge pull request #212 from chiragjangiir/fix/unclosed-quote-duplication
fix(normalization): stop unclosed quotes from duplicating the text tail
2026-10-15 07:23:37 -07:00
Chirag Jangir 08b2477e1e fix(normalization): stop unclosed quotes from duplicating the text tail
An opening quote without a matching closing mark made
_normalize_all_caps_quotes append the remainder twice: the unmatched
branch appends text[index:] and breaks without advancing index, so the
tail block below the loop appends the same remainder again. Because
emit_text normalizes every chunk before synthesis, the duplicate was
spoken aloud and written into the caption cues (issue #196). Sentence
and paragraph chunking routinely produce unclosed marks, so this hit
every quotation spanning multiple sentences.

Consume the remainder before breaking, and add regression tests that
cover the unmatched-remainder cases across quote styles plus a pair
that must still get caps normalization before an unclosed tail.

🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
2026-10-13 21:29:07 +05:30
Hari Krishna Gaddipati 2ea1941838 fix(audio_sink): fail fast when the merged-output ffmpeg encode fails
open_audio_sink() ran the merged-output ffmpeg with stderr=DEVNULL and never
checked the return code, so a failed encode silently produced a truncated file
(no moov atom) that then broke the metadata-embed step with a misleading
'exit 254'. This was observed live on a real PDF->M4B job.

Now:
- ffmpeg stderr is written to <out>.ffmpeg.log (diagnosable after the fact).
- a non-zero exit raises RuntimeError with the stderr tail and removes the
  truncated output, instead of silently closing.

Regression tests:
- a non-zero ffmpeg exit raises (not silently closes).
- ffmpeg stderr is redirected to a real file handle (not DEVNULL).

Verified: 19/19 tests in tests/test_domain_audio_sink.py pass (Python 3.11).
Also verified end-to-end on the build host: the instrumentation is what surfaced
the real embed-step ffmpeg output (all chapters mapped, moov moved) confirming a
full, chaptered 72.9 MB M4B.

Closes the diagnostic gap from homelab issue #110.
2026-10-11 18:25:52 -05:00
Deniz Şafak 315840309f Merge pull request #210 from hari2m/fix/195-mutagen-chapter-warning
fix(exporters): stop crying wolf when mutagen lacks a writable chapter API
2026-10-11 11:57:13 -07:00
Hari Krishna Gaddipati f81f2a0335 test(exporters): cover _apply_m4b_chapters_mutagen log behavior (regression for #195)
Adds TestApplyM4bChaptersMutagen to tests/test_exporters.py, covering the
method that the #195 fix touches:

- empty chapters -> returns False, logs nothing.
- mutagen installed but no writable chapter API (the expected case) ->
  returns False and logs an accurate debug note, NOT the misleading
  'mutagen is not installed' warning. This is the #195 regression.
- mutagen genuinely missing -> returns False and keeps the warning
  (backward-compatible).

Verified: the full pytest suite passes identically with and without the
fix (1494 -> 1497 passing, the delta is exactly these 3 new tests; the
pre-existing failures are environment-only and unchanged).
2026-10-11 11:36:47 -05:00
Hari Krishna Gaddipati 3fe3ac7501 fix(exporters): stop crying wolf when mutagen lacks a writable chapter API
Resolves #195.

_apply_m4b_chapters_mutagen() caught ImportError and always logged
"mutagen is not installed". That is misleading: mutagen IS installed
(>=1.47 is a hard requirement), but current releases expose no writable
MP4 chapter API - MP4Chapter (singular) does not exist, so the import
can never succeed. The scary warning implied the output was unchaptered,
when in fact ffmpeg has already written the chpl/chap chapter atoms
(and a timed-text chapter track) earlier in embed_m4b_metadata.

Distinguish the two failure modes:
- mutagen genuinely missing -> keep the warning (unchanged).
- mutagen present but no writable chapter API (the expected case) ->
  demote to a debug log with an accurate explanation that the chapters
  are already in place via ffmpeg.

No behavioral change to the output: the method still returns False in
both cases; only the log level/message of the expected path changes.
2026-10-11 10:01:16 -05:00
Mohammad Hijjawi dc16b4e418 test(executor): pin repeat-conversion output path and source-stem fallback
Run execute_conversion on a plan from build_conversion_plan and check
that an existing book.wav stays untouched while the new output goes to
book_2.wav, that a third run gets book_3.wav, that project mode checks
its audio subfolder, and that a request without original_filename is
named after the source file. Note both changes in CHANGELOG.md.
2026-10-11 06:56:13 +01:00
Mohammad Hijjawi 08acd8e8fd fix(output): use collision-free merged path so repeat conversions don't overwrite
resolve_output_layout computed a unique base name via resolve_unique_path
but discarded it, and the executor rebuilt the merged audio path from the
raw base name. Converting the same book twice silently overwrote the
previous audiobook (and its subtitles, which derive from the audio path).

- store the unique path in OutputLayout.merged_path (checked against the
  directory the audio is actually written to)
- make resolve_merged_path prefer layout.merged_path
- have the executor use resolve_merged_path instead of its own _base_name
- add a regression test

Fixes #203
2026-09-25 21:47:47 +01:00
Deniz Şafak 08e2ee8b85 fix(normalization): improve punctuation handling and spacing in text normalization
fix(subtitles): enhance ellipsis and paragraph break handling in subtitle processing
feat(gui): implement background update check for new versions
test(tests): add tests for ellipsis handling and paragraph breaks preservation
2026-09-07 17:29:27 +03:00
Deniz Şafak be74c69507 fix(subtitles): fix quotation mark spacing corruption and quoted dialogue splitting
- kokoro_text_normalization:
  - refactor _cleanup_spacing to contextually distinguish opening vs closing
    straight quotes (" and ') so leading quotes do not have spaces added after
    them and spaces before opening quotes are preserved
  - recognize non-English opening delimiters (¡, ¿, «, 「, etc.) and closing
    delimiters (», 」, etc.) in spacing normalization
  - in normalize_apostrophes, reconstruct text using original token offsets
    to preserve inter-token whitespace instead of blindly joining with spaces
- subtitle_generation:
  - add _is_sentence_boundary to recognize sentence punctuation followed by
    closing quotes/brackets (.", !", ?") in regex and karaoke modes
  - support proportional splitting for single FakeToken multi-sentence segments
    in spaCy mode (Supertonic / non-English Kokoro)
  - support newline splitting for Line mode in FakeToken fallbacks
  - safely convert language and subtitle_mode inputs
- tests:
  - add tests/test_subtitle_scenarios.py covering quotes, dialogues,
    paragraphs, contractions, all subtitle modes, and all TTS token styles
2026-08-29 12:45:18 +03:00
Deniz Şafak ffac4a4da9 fix(subtitles): enable "N words" and karaoke modes in PyQt
Three related fixes:

- make_subtitle_writer: accept word-count modes ("N words") in
  create_subtitle_writer. SubtitleMode("5 words") raised ValueError,
  which make_subtitle_writer swallowed and returned None, so PyQt never
  created or wrote the subtitle file when an "N words" mode was selected.
  Unknown modes now fall back to SubtitleMode.SENTENCE (writers only
  branch on SENTENCE_HIGHLIGHT). Regression test added.

- gui.py: refresh subtitle combo item availability after initial
  voice/profile selection. update_subtitle_options_availability() ran
  during initUI with selected_lang=None for a profile, taking the
  non-English branch and disabling Highlighting/N-words items; the fix-up
  never re-ran because setCurrentIndex on an already-current index emits
  no signal. Now called once more after the profile/voice language is
  resolved.

- AssWriter: stop discarding per-word karaoke timing. _add_karaoke_tags
  unconditionally replaced entry text with uniform {\\k100} tags,
  destroying the real per-word {\\kf} timings from
  _process_karaoke_highlighting. Only synthesize simplified tags when the
  text has no karaoke tags. Regression test added.
2026-08-20 23:27:49 +03:00
Deniz Şafak 5432de7ac5 fix(segmentation): process TTS segments per sentence, fix all subtitle modes
Sentence modes processed all text as a whole: Pipeline.__call__ merged every
engine segment back into one (whole text, no per-token timings), producing a
single giant subtitle and whole-text progress logs.

- tts_plugin/types: add TokenTiming, AudioSegment, SynthesizedAudio.segments
- tts_plugin/utils: Pipeline yields one Segment per engine segment (with
  tokens); merged fallback only when engine provides none
- kokoro engine: expose per-segment graphemes/audio + per-word token timings
- supertonic engine: expose per-segment graphemes/audio (no tokens)
- split_pattern: English Sentence/Sentence+Comma engine split is newline-only
  (boundaries applied at subtitle time via spaCy); non-English Sentence+Comma
  with spaCy ON uses spaCy pre-segmentation + newline engine split (no
  commas); spaCy-off fallback keeps comma pattern
- tts_segments: restore inter-segment whitespace on real per-word token
  boundaries only (never FakeToken fallbacks)
- _to_language_enum: accept Language enum input (str(enum) is "Language.ES",
  silently resolved to EN_US and disabled spaCy pre-TTS for every language
  in WebUI)
- pyqt/conversion, utils: replace print with logging
- add AGENTS.md documenting the segmentation/subtitle contract for future
  sessions
- tests: update English split-pattern expectations (1566 passing)
2026-08-20 23:00:15 +03:00
Deniz Şafak 823f5be029 fix(pyqt): make app close fast and crash-free
Clicking the window close button froze the app: closeEvent ran the full
process cleanup (engine disposal, CUDA flush, subprocess termination) and
unbounded thread joins synchronously on the GUI thread before the window
could start closing, and a 3.5s delay came from flush_cuda importing
torch even when it was never loaded.

- closeEvent no longer runs cleanup synchronously; the aboutToQuit hook,
  which was never connected (registered before QApplication existed), now
  runs it after the window is gone
- bound thread waits in cleanup_conversion_thread/cleanup_preview_threads
  with a terminate() fallback so closing never hangs
- flush_cuda skips torch work when torch was never imported
- restore the default Qt message handler before sys.exit; the custom
  Python handler was invoked during interpreter teardown and caused a
  SIGSEGV after shutdown cleanups finished
- log and time the close sequence (closeEvent steps + each shutdown hook)
2026-08-20 22:00:08 +03:00
Deniz Şafak aaa6ac112b fix(audio_helpers): update opus bitrate to 128kbps in ffmpeg command 2026-08-20 21:32:07 +03:00
Deniz Şafak 11274ad6bf fix(pyqt): migrate GUI to Language enums, fix subtitle mode gating
Complete the Language enum unification (commit 0dc491e) for the PyQt
GUI, which was left resolving languages from kokoro letter codes
(voice[0]) while domain and WebUI already used Language enums.

- selected_lang is now always a Language enum: voices resolve via
  language_for_voice_id(), profiles via resolve_profile_language()
- kokoro letter->Language mapping stays in the engine (new public
  language_for_code()); the Language.from_code() domain shim is removed
- legacy profile files with letter codes ("a", "e", ...) are tolerated
  only at the profile read boundary; new profiles save ISO codes
- conversion.py letter comparisons replaced with enum comparisons, and
  SynthParams lang_code= -> language= (would TypeError at runtime)
- subtitle dropdown enabled for all languages; word-count and
  highlighting modes restricted to English with auto-switch to Sentence
- tests: engine mapping + profile language resolution added; 1566 pass
2026-08-20 21:28:30 +03:00
Deniz Şafak f340b976db feat: added colorful logging using rich, improved startup times by lazy-loading spacy, some other fixes 2026-08-20 20:52:08 +03:00
Deniz Şafak 9da15aefa4 Merge pull request #192 from Sanjays2402/fix/preserve-user-metadata
Preserve user-entered book metadata after extraction
2026-08-20 09:37:34 -07:00
Deniz Şafak 919aea9295 Merge pull request #199 from SimonFoobar648/fix/star-history-chart
Fix broken star history chart
2026-08-16 22:44:10 +03:00
SimonFoobar648 ce1fc0c880 fix: fix broken star history chart
The star history chart in the README was broken. Point it to the new domain so the chart renders again.
2026-08-16 11:33:00 +00:00
Artem Akymenko 94e6b3f62e refactor: move calculate_text_length to domain, add build_chapter_payload to app-layer
- Create domain/text_utils.py with canonical calculate_text_length
  (strips chapter markers, voice markers, metadata tags)
- Remove calculate_text_length from utils.py and subtitle_utils.py
- Update all imports to use domain.text_utils directly
- Create application/chapter_selection.py with build_chapter_payload()
  (orchestrates preselection + character count + safety net)
- Replace manual chapter logic in form.py with build_chapter_payload()
- 22 new tests for text_utils and chapter_selection (1550 total)
2026-07-30 14:57:10 +00:00
Artem Akymenko d334266238 refactor: remove eager voice cache bootstrap from WebUI
Voice cache is now initialized lazily by domain's initialize_voice_cache()
in PipelinePool.get() — downloads only needed voices per request instead
of all ~100 kokoro voices at service startup. Reduces redundant downloads
and unifies initialization path for all UIs.
2026-07-30 13:45:20 +00:00
Artem Akymenko 2d18501839 refactor: consolidate metadata normalization into domain layer
- Add expand_metadata_aliases() for concept fan-out (series→5, author→2,
  description→2, tags→3 keys)
- Replace _normalize_metadata_tags in webui/service.py with
  normalize_metadata_map from domain
- Replace _normalize_metadata in epub3/exporter.py with
  normalize_metadata_map from domain
- Replace manual fan-out in form.py with expand_metadata_aliases()
- Rewrite metadata_overrides.py to use expand_metadata_aliases()
- Remove unused normalize_metadata_casefold import from exporters.py
- 19 new tests for expand_metadata_aliases (1528 total)
2026-07-30 13:04:44 +00:00
Artem Akymenko 464bf8e17d refactor: move speaker metadata logic to domain layer
Move build_narrator_roster, build_speaker_roster, match_configured_speaker,
apply_speaker_config_to_roster, prepare_speaker_metadata from
webui/routes/utils/voice.py to domain/speaker_metadata.py.

- prepare_speaker_metadata now accepts optional inject_recommended callback
  for UI-specific voice enrichment
- 47 new tests in test_domain_speaker_metadata.py
- All 1509 tests pass
2026-07-30 12:00:33 +00:00
Artem Akymenko 9bf4f8e809 feat: add voice resolution to domain layer, migrate WebUI imports
- Add formula_from_profile, resolve_profile_voice, resolve_voice_setting, resolve_voice_choice to domain/voice_resolution.py
- Add build_voice_catalog, filter_voice_catalog to domain/voice_catalog.py
- Update webui/routes/utils/voice.py to import from domain
- Update webui/routes/utils/form.py and voices.py to import from domain directly
- Update synthesize.py to use domain resolve_voice
- Add 29 tests for voice resolution functions
2026-07-30 10:58:23 +00:00
Artem Akymenko f802fb2af6 refactor: centralize integration config in domain layer, fix OutputFormat enum serialization
- Move stored_integration_config(), build_audiobookshelf_config() to domain/settings_core.py
- Remove legacy fallback from stored_integration_config() (only config[integrations])
- Add load_audiobookshelf_config() as combined entry point
- PostConversionHooks reads config directly via stored_integration_config()
- WebUI imports from domain/settings_core instead of webui/routes/utils/settings
- Remove duplicate _build_abs_config() from PostConversionHooks
- Remove dead audiobookshelf code from infrastructure/exporters.py
- Remove duplicate audiobookshelf functions from webui/service.py
- Fix OutputFormat enum serialization in output_layout_service.py and conversion_executor.py
  (f'{enum}' gave 'OutputFormat.WAV' instead of '.wav')
- Mock spaCy in test_returns_at_least_one_segment instead of loading real model
- Add 28 tests for PostConversionHooks and build_audiobookshelf_config
2026-07-29 13:11:56 +00:00
Artem Akymenko c293cc90f6 chore: suppress phonemizer word-count-mismatch warnings (normal behavior) 2026-07-29 11:26:43 +00:00
Artem Akymenko 696ce1ebd0 fix: use existing to_float32 from audio_helpers, keep INFO log level
- synthesize.py: import to_float32 from domain.audio_helpers instead of duplicate
- Revert all log levels back to INFO (user controls verbosity via --log-level)
2026-07-29 11:24:01 +00:00
Artem Akymenko d3ded8af0e feat: add comprehensive logging throughout conversion pipeline
- voice_resolver: log spec resolution and cache hits
- conversion_executor: log chapter start/end, voice resolution, timing
- conversion_planner: log plan summary
- conversion_service: log entry point params and outcome
- conversion_runner: log job params and lifecycle
- form.py: log PendingJob creation
- synthesize.py: log pipeline creation and device selection
2026-07-29 11:05:52 +00:00
Artem Akymenko c706f7714a fix: language enum coercion, preview logging, file picker
- form.py: convert language string to Language enum at all entry points
- conversion_request.py: add _coerce_enums() for defense-in-depth
- synthesize.py: fix NameError (lang -> language) in preview pipeline
- voice.py: render LANGUAGE_DESCRIPTIONS keys as .value strings for Jinja
- dashboard.js: open file picker on dropzone click
- api.py: add logging to preview endpoint for debugging
2026-07-29 10:57:21 +00:00
Artem Akymenko 2b70b9ca45 feat: Supertonic language + total_steps propagation
Language enum expanded from 9 to 33 languages:
- Added 24 new ISO 639-1 languages: AR, BG, CS, DA, DE, EL, ET, FI,
  HR, HU, ID, KO, LT, LV, NL, PL, RO, RU, SK, SL, SV, TR, UK, VI
- Updated display_name, is_cjk (added KO)

Supertonic language mapping (32 languages, no ZH):
- engine.py: _SUPERTONIC_LANG_MAP, engine_language(), supported_languages()
- __init__.py: create_engine() passes config.language to pipeline
- pipeline.py: __init__() accepts language, resolves to ISO code;
  __call__() passes lang= to TTS.synthesize()

total_steps propagation:
- tts_segments(): +total_steps param, conditionally passed to backend
- synthesize_text(): +total_steps param
- run_tts_segment_loop(): +total_steps param
- executor: all 5 synthesize_text() calls pass total_steps

Integration:
- pipeline_factory: create_pipeline_for_job() passes language to supertonic
- preview path: create_pipeline('supertonic', language=language)

Tests updated to accept total_steps in FakeBackend.__call__
2026-07-28 14:09:42 +03:00
Artem Akymenko 953bef1e71 refactor: group ConversionRequest fields into config objects
Domain config types (domain/config_types.py):
- PronunciationConfig: pronunciation/heteronym/normalization overrides
- SubtitleConfig: mode, format, max_words
- CoverConfig: path, mime

Domain functions now accept config objects:
- build_tts_context(subtitle=, pronunciation=) instead of 9 individual params
- make_subtitle_writer(subtitle=) instead of 3 params
- process_and_write_subtitles(subtitle=) instead of 2 params
- embed_m4b_metadata(cover=) instead of 2 params
- build_epub3_package(cover=) instead of 2 params

ConversionRequest: 18 flat fields + 8 config objects
Application/config.py re-exports domain types
All tests updated to new API
2026-07-28 13:41:45 +03:00
Artem Akymenko 146cc81271 refactor: centralize cleanup in app layer
- New: application/cleanup.py — flush_cuda(), dispose_engines(), cleanup(), register_ui_cleanup()
- conversion_service.py finally: pool.dispose_all() + voice_cache.clear() + flush_cuda()
- webui/conversion_runner.py: removed gc/cuda finally block (cleanup in run_conversion)
- shutdown.py: 160→120 lines, 5 inline cleanups → 4 process-level + app_cleanup() delegation
- Fixed bugs: _PIPELINES (didn't exist), PluginManager.dispose_all() (never called),
  VoiceCache.clear() (never called in finally), duplicate cleanup removed
2026-07-28 13:41:28 +03:00
Artem Akymenko 61204cc389 refactor: move pool/cache/resolver into run_conversion()
- New: application/voice_resolver.py — AppVoiceResolver (app-layer, takes ConversionRequest)
- conversion_service.py: run_conversion(request, events) creates PipelinePool, VoiceCache, AppVoiceResolver internally
- conversion_runner.py: 268→165 lines, removed WebUIVoiceResolver, pool/cache creation
- Tests: mock fixture for pool/resolver, removed unused variables
- Added pool.dispose_all() in finally block for cleanup
2026-07-27 13:37:19 +03:00