Dymas/abogen
1
0
Fork 0
mirror of https://github.com/denizsafak/abogen.git synced 2026-07-18 13:40:27 +02:00
Code Issues Packages Projects Releases Wiki Activity
695 Commits 5 Branches 27 Tags
096ea58d742f1adf2ec3061275f1edad52dc86da
Commit Graph
5 Commits
Author SHA1 Message Date
Artem Akymenko c094b94704 feat(tts-plugin): complete Plugin Architecture refactor 
- Normalize Pipeline public API: create_pipeline(plugin_id, *, lang_code, device)
- EngineConfig: add lang_code field per Architecture Amendment #1
- Kokoro plugin reads config.lang_code (fixes functional regression)
- Static voice catalog in PluginManifest.voices (None = dynamic/VoiceLister)
- get_voices() reads from manifest without creating Engine
- Remove dead kwargs (sample_rate, auto_download, total_steps) from SuperTonic
- Clean up unused imports and dead code in engine implementations
- Fix test expectations for VoiceLister (mock overrides)
- Add clear_preview_pipelines() for resource management
 2026-07-12 16:20:20 +03:00
Artem Akymenko 735098d7cd feat: add static voice catalog to PluginManifest 
- Add  to PluginManifest
  - None = not declared (use VoiceLister fallback)
  - () = explicitly no static voices
  - Non-empty = static catalog available without Engine instantiation

- Update get_voices() to check manifest first, fall back to Engine
- Declare 54 Kokoro voices and 10 SuperTonic voices in manifests
- Remove hardcoded voice lists from engine.py files
- Engine.listVoices() now returns [] (manifest is source of truth)

- Clean up dead create_pipeline() kwargs (sample_rate, auto_download, total_steps)
  - SuperTonic plugin uses internal defaults
  - total_steps is per-request parameter via Pipeline.__call__() kwargs

- Add clear_preview_pipelines() for resource cleanup
- Fix test mocks to override listVoices()
- Update Architecture Amendment #1 doc
 2026-07-12 16:20:16 +03:00
Artem Akymenko 9150a80459 refactor: eliminate remaining legacy dependencies from production code 
Task 1: Replace hardcoded VoiceLister bypass in get_voices()
- Use PluginManager → Engine → VoiceLister instead of direct imports
- No more hardcoded imports of plugins.kokoro.engine / plugins.supertonic.engine

Task 2: Remove SuperTonic Plugin dependency on legacy backend
- Create self-contained plugins/supertonic/pipeline.py
- Plugin no longer imports from abogen.tts_backends

Production code now has zero imports from:
- abogen.tts_backend
- abogen.tts_backend_registry
- abogen.tts_backends
 2026-07-12 16:20:06 +03:00
Artem Akymenko 6284c501ed feat: add SuperTonic TTS plugin 
- Add plugins/supertonic/ with Engine and EngineSession implementations
- Reuse existing SupertonicPipeline from abogen.tts_backends.supertonic
- Implement VoiceLister capability (M1-M5, F1-F5 voices)
- Declare no streaming support via capabilities
- Add 28 tests: plugin loading, protocol compliance, lifecycle, voice listing, parameters, errors
- All 237 tests pass
 2026-07-12 16:20:06 +03:00
Artem Akymenko d129b0abe8 feat: add Kokoro plugin vertical slice 
Implement first TTS plugin using new Plugin Architecture:
- plugins/kokoro/: Plugin package with manifest and entry point
- plugins/kokoro/engine.py: KokoroEngine and KokoroSession adapters
- Wraps existing KokoroBackend without modifying it
- Implements VoiceLister capability
- Satisfies Engine/EngineSession protocol
- Passes all 163 contract tests

Tests:
- Plugin loading through Plugin Loader
- Manifest validation
- Engine creation and lifecycle
- Session synthesis and dispose
- VoiceLister capability

15 new tests for Kokoro plugin.
 2026-07-12 16:20:05 +03:00