yt-dbl 1.0.0__tar.gz

yt_dbl-1.0.0/.coveragerc

@@ -0,0 +1,14 @@
+[run]
+source = yt_dbl
+branch = true
+
+[report]
+show_missing = true
+skip_empty = true
+fail_under = 80
+exclude_lines =
+    pragma: no cover
+    if TYPE_CHECKING:
+    if __name__
+    @overload
+    \.\.\.

yt_dbl-1.0.0/.env.example

@@ -0,0 +1,4 @@
+YT_DBL_ANTHROPIC_API_KEY=
+YT_DBL_HF_TOKEN=
+YT_DBL_SEPARATION_SEGMENT_SIZE=256
+YT_DBL_SEPARATION_BATCH_SIZE=1

yt_dbl-1.0.0/.github/workflows/ci.yml

@@ -0,0 +1,68 @@
+name: CI
+
+on:
+  push:
+    branches: [master]
+  pull_request:
+    branches: [master]
+  workflow_call:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  UV_TORCH_BACKEND: cpu
+
+jobs:
+  quality:
+    name: Lint & Type Check
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: uv.lock
+
+      - name: Install dependencies
+        run: uv sync --python 3.12 --frozen
+
+      - name: Ruff lint
+        run: uv run ruff check src/ tests/
+
+      - name: Ruff format
+        run: uv run ruff format --check src/ tests/
+
+      - name: Mypy
+        run: uv run mypy src/ tests/
+
+  test:
+    name: Test
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: astral-sh/setup-uv@v5
+        with:
+          enable-cache: true
+          cache-dependency-glob: uv.lock
+
+      - name: Install dependencies
+        run: uv sync --python 3.12 --frozen
+
+      - name: Run unit & integration tests
+        run: >-
+          uv run pytest tests/unit/ tests/integration/
+          -n auto
+          --cov --cov-report=xml --cov-report=term-missing
+          -p no:randomly
+
+      - name: Upload coverage
+        if: ${{ !cancelled() }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: coverage-report
+          path: coverage.xml
+          retention-days: 7

yt_dbl-1.0.0/.github/workflows/release.yml

@@ -0,0 +1,43 @@
+name: Release
+
+on:
+  push:
+    branches: [master]
+
+concurrency:
+  group: release
+  cancel-in-progress: false
+
+jobs:
+  ci:
+    name: CI
+    uses: ./.github/workflows/ci.yml
+
+  release:
+    name: Semantic Release
+    needs: ci
+    runs-on: ubuntu-latest
+    permissions:
+      id-token: write
+      contents: write
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.GITHUB_TOKEN }}
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - uses: astral-sh/setup-uv@v5
+
+      - name: Semantic Release
+        id: release
+        uses: python-semantic-release/python-semantic-release@v9.21.0
+        with:
+          github_token: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Publish to PyPI
+        if: steps.release.outputs.released == 'true'
+        uses: pypa/gh-action-pypi-publish@release/v1

yt_dbl-1.0.0/.gitignore

@@ -0,0 +1,14 @@
+.env
+__pycache__/
+*.pyc
+.ruff_cache/
+.pytest_cache/
+.mypy_cache/
+.coverage
+coverage.xml
+htmlcov/
+dist/
+*.egg-info/
+work/
+/models/
+.venv/

yt_dbl-1.0.0/.pre-commit-config.yaml

@@ -0,0 +1,50 @@
+repos:
+  - repo: https://github.com/astral-sh/ruff-pre-commit
+    rev: v0.15.0
+    hooks:
+      - id: ruff
+        args: [--fix]
+      - id: ruff-format
+
+  - repo: https://github.com/pre-commit/mirrors-mypy
+    rev: v1.19.1
+    hooks:
+      - id: mypy
+        additional_dependencies:
+          - pydantic>=2.12.5
+          - pydantic-settings>=2.12
+          - typer>=0.21.1
+          - rich>=14.3.2
+          - anthropic>=0.78
+          - numpy>=2.3
+          - soundfile>=0.13
+        args: [--config-file=mypy.ini]
+        pass_filenames: false
+
+  - repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v6.0.0
+    hooks:
+      - id: trailing-whitespace
+      - id: end-of-file-fixer
+      - id: check-yaml
+      - id: check-json
+      - id: check-merge-conflict
+      - id: debug-statements
+
+  - repo: https://github.com/compilerla/conventional-pre-commit
+    rev: v4.3.0
+    hooks:
+      - id: conventional-pre-commit
+        stages: [commit-msg]
+        args:
+          - feat
+          - fix
+          - refactor
+          - style
+          - build
+          - ci
+          - docs
+          - test
+          - perf
+          - chore
+          - revert

yt_dbl-1.0.0/CHANGELOG.md

@@ -0,0 +1,474 @@
+# CHANGELOG
+
+
+## v1.0.0 (2026-02-06)
+
+### Bug Fixes
+
+- Add missing __init__.py to models package
+  ([`e4c4b5b`](https://github.com/brolnickij/yt-dbl/commit/e4c4b5be5ab1a9b7ebeef2a205aa1facfd0d5c37))
+
+- Add sidechain ducking and limiter to reduce robotic artifacts in mixed audio
+  ([`ac60ce0`](https://github.com/brolnickij/yt-dbl/commit/ac60ce081a49693dd56514c2ae2a4e86b0a50f2d))
+
+- Allow lazy torch import in model manager lint config
+  ([`110a3ca`](https://github.com/brolnickij/yt-dbl/commit/110a3ca23573a55e72decb68d69a4ebb54921c13))
+
+- Correct TTS sample rate from 12kHz to 24kHz (critical audio fix)
+  ([`693e5db`](https://github.com/brolnickij/yt-dbl/commit/693e5dbdbf4643b12a657f5013ffd84524a0f409))
+
+The Qwen3-TTS model outputs audio at 24000 Hz (model.sample_rate), but we were writing WAV files at
+  12000 Hz. This caused all synthesized audio to play at half speed with pitched-down 'morphed'
+  voices.
+
+Also: - Revert TTS params to model defaults (temp=0.9, top_k=50, top_p=1.0) - Remove dead speed
+  pre-calculation logic (model.generate speed param is documented as 'not directly supported yet') -
+  Update tests to match corrected sample rate and defaults
+
+- Improve TTS quality with proper Qwen3-TTS parameters
+  ([`73a9cbb`](https://github.com/brolnickij/yt-dbl/commit/73a9cbb21631b55479bff407102003bc9f2b299e))
+
+- Set temperature to 0.6 (model default) instead of 0.9 - Add top_k=20, top_p=0.8,
+  repetition_penalty=1.05 params - Use native 'speed' param in model.generate() to pre-fit
+  translated text into original segment duration, reducing reliance on quality-degrading post-hoc
+  ffmpeg atempo - Add _estimate_speaking_rate() and _estimate_tts_speed() methods for intelligent
+  speed pre-calculation - Add unit tests for new config defaults and speed estimation
+
+- Resolve relative paths in _rename_outputs, add E2E tests
+  ([`7edc5f8`](https://github.com/brolnickij/yt-dbl/commit/7edc5f87f2d8bc20620be0a094889371ec165818))
+
+- Fix _rename_outputs to resolve relative paths against step_dir (audio-separator returns filenames,
+  not absolute paths) - Add test_e2e_separate.py: 4 E2E tests for real BS-RoFormer separation - Add
+  unit test for relative path handling in _rename_outputs - Update E2E pipeline tests: require
+  audio_separator_available fixture - CI: restrict test job to tests/unit/, remove e2e job -
+  Justfile: run E2E with -n0 to prevent concurrent model downloads
+
+All checks pass: lint ✓ format ✓ mypy ✓ 115 unit/integration ✓ 10 E2E ✓
+
+- Strip ANSI codes in CLI help test for narrow CI terminals
+  ([`b4ac57b`](https://github.com/brolnickij/yt-dbl/commit/b4ac57b3f0055ddd58f015f290d00289e8185d16))
+
+- Update environment variable names for clarity and consistency
+  ([`fdfa808`](https://github.com/brolnickij/yt-dbl/commit/fdfa8081583bad2b796522fee69ba18e2173b509))
+
+- **release**: Use python -m build in semantic-release Docker container
+  ([`0ad5354`](https://github.com/brolnickij/yt-dbl/commit/0ad5354a3112c02ec4ff0b8b72a8f3cdd87c388b))
+
+### Build System
+
+- Add conventional commits enforcement via pre-commit
+  ([`fbcd057`](https://github.com/brolnickij/yt-dbl/commit/fbcd057c245395cec621c8a20618a866d441364c))
+
+- Add justfile for development tasks
+  ([`3a056f8`](https://github.com/brolnickij/yt-dbl/commit/3a056f8543b0b2817a8c17e3229cf4d8632ff8fc))
+
+- Bump all dependencies to latest versions
+  ([`733472a`](https://github.com/brolnickij/yt-dbl/commit/733472a11f95630c8422399e524312e6d4fac8f6))
+
+- Bump all dependency minimum versions to latest
+  ([`ce66a82`](https://github.com/brolnickij/yt-dbl/commit/ce66a82257164c2e50226b2507f85b8ff9f54cdd))
+
+- anthropic: >=0.52 -> >=0.78 - rich: >=14.0 -> >=14.3.2 - pydantic: >=2.12 -> >=2.12.5 - yt-dlp:
+  >=2026.0 -> >=2026.2 - audio-separator: >=0.41 -> >=0.41.1 - mlx-audio: >=0.3 -> >=0.3.1 - typer:
+  >=0.21 -> >=0.21.1 - mypy: >=1.19 -> >=1.19.1 - pytest: >=9.0 -> >=9.0.2 - pytest-randomly: >=4.0
+  -> >=4.0.1 - pre-commit: >=4.5 -> >=4.5.1 - pre-commit hooks additional_dependencies synced
+
+- Update lockfile with new dependencies
+  ([`53a9ca8`](https://github.com/brolnickij/yt-dbl/commit/53a9ca82409c706193ea10a83be494fa37d9a614))
+
+### Chores
+
+- Add TC002 to test per-file-ignores in ruff config
+  ([`ecef19b`](https://github.com/brolnickij/yt-dbl/commit/ecef19b52c0594b1ef17332354e848f061f0366a))
+
+- Enhance pyproject.toml
+  ([`aedd66f`](https://github.com/brolnickij/yt-dbl/commit/aedd66f56a823be10e41da1362feb5b11025905e))
+
+- Remove unused network marker and dead _HAS_MLX variable
+  ([`c7e00c2`](https://github.com/brolnickij/yt-dbl/commit/c7e00c26cb41340bb9927f463d3747ddef3f093d))
+
+- Stable version
+  ([`0518d3e`](https://github.com/brolnickij/yt-dbl/commit/0518d3eee2339fdb710eac6e11b464c86f939075))
+
+- Update .gitignore with coverage and mypy cache
+  ([`7325c43`](https://github.com/brolnickij/yt-dbl/commit/7325c4309f88a2891f0de64cd78af52ed981454f))
+
+### Code Style
+
+- Use parameterless raise for typer.Exit
+  ([`51e8ca7`](https://github.com/brolnickij/yt-dbl/commit/51e8ca73b33a796943ab1909a3eefae5684a776d))
+
+### Continuous Integration
+
+- Add GitHub Actions workflow for lint, typecheck, and tests
+  ([`cf3f0d3`](https://github.com/brolnickij/yt-dbl/commit/cf3f0d348292236fac3ab117ac74a8b0a0139f5a))
+
+- Add platform marker for mlx-audio, set UV_TORCH_BACKEND=cpu in CI
+  ([`ed01b02`](https://github.com/brolnickij/yt-dbl/commit/ed01b022b1fe33d9456b292ef7d513bfb13a973e))
+
+- Drop Python version matrix in favor of single 3.12
+  ([`b9b7063`](https://github.com/brolnickij/yt-dbl/commit/b9b7063174e635ece70f1fe7282c3914f336cf7e))
+
+- Force E2E tests to run sequentially (xdist_group)
+  ([`bdd201f`](https://github.com/brolnickij/yt-dbl/commit/bdd201fcf4a55cb3cc0f9641d387168ac9e61ffc))
+
+GPU-heavy models (VibeVoice-ASR 9B, BS-RoFormer) kill the machine when loaded in parallel. All E2E
+  tests now share a single xdist group via conftest hook + --dist loadgroup in pytest.ini.
+
+- Include integration tests
+  ([`fb72eee`](https://github.com/brolnickij/yt-dbl/commit/fb72eee65eb39ec9d6b8baf26779b2f5964c1f16))
+
+### Documentation
+
+- Bump
+  ([`0812074`](https://github.com/brolnickij/yt-dbl/commit/081207411751992b400907a09b86713b0f081e27))
+
+- Bump
+  ([`10c1d62`](https://github.com/brolnickij/yt-dbl/commit/10c1d62a8edf9e28c55e1eda90a342292492b344))
+
+- Bump readme.md
+  ([`1ad90a2`](https://github.com/brolnickij/yt-dbl/commit/1ad90a2429ce33193eccc4904b67a3252c51792c))
+
+- Bump readme.md
+  ([`e2679d7`](https://github.com/brolnickij/yt-dbl/commit/e2679d775790e6f9d79ae37dd93203e9ed4f4b13))
+
+- Bump readme.md
+  ([`e8cf40e`](https://github.com/brolnickij/yt-dbl/commit/e8cf40e9bf3919e850f25c92e114359dfb2424ff))
+
+- Cleanup
+  ([`baf712e`](https://github.com/brolnickij/yt-dbl/commit/baf712ebe2da3c851ed7dd05ce38792da0bb5a40))
+
+- Cleanup
+  ([`a87505a`](https://github.com/brolnickij/yt-dbl/commit/a87505a6a3fb8d042633173da9fab3b17b055242))
+
+- Fix repository URL
+  ([`392b214`](https://github.com/brolnickij/yt-dbl/commit/392b214db46bce01fb433732e8d3c6d7235a41dd))
+
+- Improve formatting and clarity in README.md
+  ([`0cd95a7`](https://github.com/brolnickij/yt-dbl/commit/0cd95a7e3d709dacabd695678d96ea835edbfb8c))
+
+- Update README
+  ([`eb2e9ab`](https://github.com/brolnickij/yt-dbl/commit/eb2e9ab142c3cc7ce2459a91d1b3d3988c1a90d0))
+
+- Update requirements & install
+  ([`9d1d826`](https://github.com/brolnickij/yt-dbl/commit/9d1d8261d74f2ec57fcc4ddaa26f58a7c5f1117e))
+
+- **separate**: Document ModelManager exclusion for separator models
+  ([`c15487f`](https://github.com/brolnickij/yt-dbl/commit/c15487fd9edc0cd1f1a8517813af57031e114de1))
+
+### Features
+
+- Add LRU model manager
+  ([`b577721`](https://github.com/brolnickij/yt-dbl/commit/b577721e9a83d7696f3b909206dfde90cf57ea2f))
+
+- Add Rich progress bars to transcribe and synthesize steps
+  ([`ebb2d43`](https://github.com/brolnickij/yt-dbl/commit/ebb2d43bdc456c7def2a857ea02a969f56950f60))
+
+- ASR: animated spinner with status during long model.generate() call - Alignment: progress bar
+  showing segment N/M - TTS synthesis: progress bar per segment - Postprocessing: progress bar for
+  speed/loudnorm
+
+- Add stress marks to Russian translations for better TTS pronunciation
+  ([`c6b9683`](https://github.com/brolnickij/yt-dbl/commit/c6b9683d3b7f1f97e0ec107f85d66d816bc0e240))
+
+- Add U+0301 combining acute accent instruction to Claude translation prompt for Russian (ru) and
+  Ukrainian (uk) target languages - Stress marks affect Qwen3-TTS tokenization, improving word
+  stress and pronunciation naturalness - Disable ruff RUF001/RUF003 (ambiguous Unicode) — Cyrillic
+  text is intentional - Zero extra dependencies or API calls (reuses existing Claude translation
+  call)
+
+- Confidence-weighted voice reference selection for better cloning
+  ([`27f9738`](https://github.com/brolnickij/yt-dbl/commit/27f97389331f1343cc27233121230761d953d2ec))
+
+- De-ess sibilants in TTS postprocessing chain
+  ([`334bb09`](https://github.com/brolnickij/yt-dbl/commit/334bb0959ad3ce1394e0e4f31422ddd627cedd2c))
+
+- Enhance TTS guidelines for better pronunciation and clarity
+  ([`6688b45`](https://github.com/brolnickij/yt-dbl/commit/6688b4520eb27b0e211f8ede72d40b4d2d090c35))
+
+- Equal-power crossfade for constant energy at segment boundaries
+  ([`03ed67a`](https://github.com/brolnickij/yt-dbl/commit/03ed67a3825a262b67e2221b4892886a1ea04361))
+
+- Highpass + denoise filter on voice reference for cleaner cloning
+  ([`393815f`](https://github.com/brolnickij/yt-dbl/commit/393815fee7e43d1cef0eb9b080eba92b51b0d642))
+
+- Implement real vocal separation with BS-RoFormer
+  ([`8c394f0`](https://github.com/brolnickij/yt-dbl/commit/8c394f0ee7bb85f4bbb0842f8d031d1ca49b205e))
+
+- Replace stub SeparateStep with audio-separator integration - Use BS-RoFormer (SDR 12.97) as
+  default model - CoreML acceleration for Apple Silicon (M4 Pro) - Configurable model, segment_size,
+  overlap via Settings - Idempotency: skip if outputs already exist - Memory cleanup: gc.collect() +
+  GPU cache clear after separation - 17 new unit tests for validation, rename, config, error paths -
+  Updated pipeline integration tests with mock separation - Added prefill_separate helper to
+  conftest
+
+- Implement real yt-dlp download step with audio extraction
+  ([`2609176`](https://github.com/brolnickij/yt-dbl/commit/260917670e2fba86b3c72ef793b5ea2be0de9585))
+
+- Implement single-pass translation via Claude Opus 4.6 (iter 4)
+  ([`b78ed00`](https://github.com/brolnickij/yt-dbl/commit/b78ed004c20d7a85b32ff1276d29a3441fb13485))
+
+- Add anthropic>=0.52 runtime dependency - TranslateStep: single API call with all segments, JSON
+  output - System prompt for dubbing-aware translation (concise for TTS timing) - Idempotent: caches
+  translations.json, regenerates subtitles.srt - Parse Claude response with markdown fence handling
+  - Config: claude_model='claude-opus-4-6', removed translation_batch_size - 22 unit tests
+  (validation, parsing, SRT, persistence, mocked API) - Integration tests updated with
+  _pipeline_patches helper - 168 tests, 91.92% coverage, all quality checks pass
+
+- Implement transcription with VibeVoice-ASR + ForcedAligner (iter 3)
+  ([`decd55c`](https://github.com/brolnickij/yt-dbl/commit/decd55c08ffec0c327d68dd5123e8329842f1275))
+
+Two-model pipeline running on Apple Silicon MLX Metal: 1. VibeVoice-ASR (9B) — ASR + diarization +
+  segment timestamps 2. Qwen3-ForcedAligner (0.6B) — word-level forced alignment
+
+Key changes: - Add mlx-audio>=0.3 dependency (requires prerelease transformers) - Add transcription
+  config: asr_model, aligner_model, max_tokens, temperature - Rewrite TranscribeStep from stub to
+  full implementation: - _run_asr(): loads VibeVoice, runs generate(), normalises key variants -
+  _run_alignment(): per-segment forced alignment with error fallback - _detect_language(): Unicode
+  heuristics (ja before zh for kana) - _extract_speakers(): duration aggregation + best reference
+  segment - Idempotent with JSON cache, gc.collect() between models - 146 tests (25 new for
+  transcription), 90.93% coverage - All quality checks pass: ruff lint, ruff format, mypy strict
+
+- Implement TTS synthesis with Qwen3-TTS voice cloning (iter 5)
+  ([`c4330e5`](https://github.com/brolnickij/yt-dbl/commit/c4330e528fa25489abc6277e574f0fc319d5760a))
+
+- Implement SynthesizeStep: voice reference extraction, TTS generation, speed adjustment (ffmpeg
+  atempo), loudness normalization (-16 LUFS) - Model: mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16
+  via mlx-audio - Voice cloning via ref_audio + ref_text params - Add tts_model, tts_temperature,
+  tts_sample_rate to Settings - Add synth_path, synth_speed_factor to Segment schema - Add
+  soundfile>=0.13 dependency for WAV writing - Add 20 unit tests for synthesize step (all passing) -
+  Update integration tests with TTS mocks - Add prefill_synthesize helper to conftest - 188 tests,
+  91.14% coverage
+
+- Init
+  ([`68eb169`](https://github.com/brolnickij/yt-dbl/commit/68eb169be5ab510a00d1db0913f20085d89219d1))
+
+- Init assemble module
+  ([`e88b499`](https://github.com/brolnickij/yt-dbl/commit/e88b499c4e4ec148cae228958b431f762ecbc01e))
+
+- Init publish configs
+  ([`dee5cbf`](https://github.com/brolnickij/yt-dbl/commit/dee5cbfa0de6c3b4b94787b439c4cb0b6eca6b70))
+
+- Initial public release
+  ([`d112074`](https://github.com/brolnickij/yt-dbl/commit/d112074bd781c73c438380dd352ab7010a6f0761))
+
+BREAKING CHANGE: first stable release of yt-dbl
+
+- Make separation batch_size configurable via env
+  ([`396ca6c`](https://github.com/brolnickij/yt-dbl/commit/396ca6c31f9570455d8028e3c73054fda08f5a6d))
+
+Add YT_DBL_SEPARATION_BATCH_SIZE (default=1, max=16). On M4 Pro with unified memory, higher values
+  (2-4) can speed up separation.
+
+- Optimize translation prompt for TTS-friendly output
+  ([`fe50b79`](https://github.com/brolnickij/yt-dbl/commit/fe50b798137a5ed75140115c284c18a5768f76b5))
+
+- Add rules: numbers→words, expand abbreviations, spoken style, no special chars - Fix rule
+  numbering bug (duplicate '7.' when stress rule injected) - Add stress-marked example in JSON
+  output for ru/uk languages - Rephrase prompt intro to emphasize TTS optimization context
+
+- Sinc resampling via scipy to eliminate aliasing artifacts
+  ([`9b9886f`](https://github.com/brolnickij/yt-dbl/commit/9b9886f52e437ae3be36a120525c040d501b628e))
+
+- Suppress HuggingFace/transformers noise during model loading
+  ([`34671fa`](https://github.com/brolnickij/yt-dbl/commit/34671fa191aac3cca65a32e9ee249b68cb98a34d))
+
+- Add suppress_library_noise() context manager to logging utils - Silence HF Hub download progress
+  bars (HF_HUB_DISABLE_PROGRESS_BARS) - Suppress 'model of type' and tokenizer regex warnings - Mute
+  noisy loggers: huggingface_hub, transformers, tqdm, etc. - Wrap all 3 model loads: ASR, aligner,
+  TTS
+
+- Update TTS mock functions for improved testing and remove mlx dependency
+  ([`b929fa8`](https://github.com/brolnickij/yt-dbl/commit/b929fa891b841fcfdeb0f4b51518a0bda315d360))
+
+- Use ffmpeg-full with rubberband for higher quality audio
+  ([`feac9a9`](https://github.com/brolnickij/yt-dbl/commit/feac9a9662864158848af40592d2c62835477fb6))
+
+- Auto-detect ffmpeg-full (brew keg-only) with fallback to system ffmpeg - Rubberband
+  pitch-preserving time-stretch instead of atempo - Two-pass loudnorm (measure → apply with
+  linear=true) - Voice references at 24kHz (Qwen3-TTS native rate, was 16kHz) - AAC output at
+  320kbps (was default ~128kbps) - Add ffmpeg_path config setting for custom ffmpeg location - Tests
+  for _detect_ffmpeg, _detect_ffprobe, has_rubberband
+
+- **cli**: Implement real models list and download commands
+  ([`b336cbf`](https://github.com/brolnickij/yt-dbl/commit/b336cbfed7f4edf94ea47e30bf2b5445d857f412))
+
+- **config**: Auto-detect max_loaded_models based on system RAM
+  ([`38ec78f`](https://github.com/brolnickij/yt-dbl/commit/38ec78f2122e55e6466bce203a8627bf08052570))
+
+- **logging**: Add memory measurement and enhanced model load/unload logging
+  ([`7db825b`](https://github.com/brolnickij/yt-dbl/commit/7db825bc41c07af2b166d6f29ad527abe8aa6686))
+
+- **manager**: Add memory tracking and MLX/torch cache cleanup
+  ([`9d3f746`](https://github.com/brolnickij/yt-dbl/commit/9d3f746db319a15cf763530b5627ee35ad3d1f96))
+
+- **models**: Export public API from models package
+  ([`5893660`](https://github.com/brolnickij/yt-dbl/commit/5893660e5df255ee25629aa6336cf1435b9a6539))
+
+- **pipeline**: Accept ModelManager in PipelineStep base class
+  ([`740dd96`](https://github.com/brolnickij/yt-dbl/commit/740dd96b33bce7ad07a66ed4343a30886099cb47))
+
+- **registry**: Add ML model registry with download status checks
+  ([`2a3e8f1`](https://github.com/brolnickij/yt-dbl/commit/2a3e8f1d1332967c34d8db5c0e077f52b24f3217))
+
+- **runner**: Wire ModelManager and memory status into pipeline runner
+  ([`34795ac`](https://github.com/brolnickij/yt-dbl/commit/34795ac7ad5b600a2fbc66f185ccdbb94f24339e))
+
+- **schemas**: Add source_language field to PipelineState
+  ([`e00b27a`](https://github.com/brolnickij/yt-dbl/commit/e00b27ab871f677c7f683e5cb96327efbc8bf838))
+
+- **synthesize**: Integrate ModelManager for TTS model lifecycle
+  ([`0cca551`](https://github.com/brolnickij/yt-dbl/commit/0cca551451173f82c81d6766b33250a3fca304d7))
+
+- **transcribe**: Integrate ModelManager and expand language detection
+  ([`2029b71`](https://github.com/brolnickij/yt-dbl/commit/2029b7109e5048ed791b5171bd4be48860e718bd))
+
+- **translate**: Pass source language to Claude translation prompt
+  ([`93376bb`](https://github.com/brolnickij/yt-dbl/commit/93376bb40dc8824629e1c9be7fd3def22edef82d))
+
+### Performance Improvements
+
+- Cache has_rubberband, unload ASR before aligner, precompute lookups, compact JSON, cleanup
+  intermediates
+  ([`8422144`](https://github.com/brolnickij/yt-dbl/commit/842214483d4ab75c25e9a73718c0e3dc6d1e36c2))
+
+- Enhance RAM-based auto-detection for `max_loaded_models` and `separation_batch_size`
+  ([`94f479a`](https://github.com/brolnickij/yt-dbl/commit/94f479af98612efed509ddfeced8bd96bfb44ab1))
+
+- Enhance segment postprocessing with parallel execution
+  ([`9d1e296`](https://github.com/brolnickij/yt-dbl/commit/9d1e2962a6568dba7b9f6dc088f5410d5e3a4f0c))
+
+- **synthesize**: Merge ffmpeg postprocessing into minimal-pass pipeline
+  ([`6db80ef`](https://github.com/brolnickij/yt-dbl/commit/6db80ef75ec61140923d3462c96cf85b2dc51521))
+
+- **synthesize**: Use np.asarray instead of tolist for mlx-to-numpy conversion
+  ([`c6d2bc4`](https://github.com/brolnickij/yt-dbl/commit/c6d2bc469bef586ada386add7d4fb0af6b0ae3f8))
+
+- **transcribe**: Load audio once and slice per segment for alignment
+  ([`9586e6d`](https://github.com/brolnickij/yt-dbl/commit/9586e6d9f0ef146d3e99c3b24c75fc08e9078022))
+
+### Refactoring
+
+- Add __all__ exports to public modules
+  ([`aff728c`](https://github.com/brolnickij/yt-dbl/commit/aff728c6b5fdbd740821e61d859853e960fbce27))
+
+- Add PEP 561 py.typed marker
+  ([`ac757cc`](https://github.com/brolnickij/yt-dbl/commit/ac757cc30cf24aac805221a74470172543cbed99))
+
+- Complete pipeline error hierarchy with TranscriptionError and AssemblyError wrapping
+  ([`4160359`](https://github.com/brolnickij/yt-dbl/commit/416035944b8d3a260a60de31839ed62ac7710fe8))
+
+- Deduplicate atempo chain in speed_up_audio
+  ([`4750e44`](https://github.com/brolnickij/yt-dbl/commit/4750e442d55153410cba53411991320693dffa65))
+
+- Deduplicate resolve_vocals and GPU cleanup into shared utilities
+  ([`ce1be33`](https://github.com/brolnickij/yt-dbl/commit/ce1be3399e521f104c2e4e7e9fc267cbe6bc9d32))
+
+- Extract audio processing utilities from SynthesizeStep into utils/audio_processing
+  ([`37ae8d1`](https://github.com/brolnickij/yt-dbl/commit/37ae8d113798009249f95fa780259bab3136c5bb))
+
+- Extract constants, fix inconsistencies, deduplicate loudnorm targets
+  ([`16c815d`](https://github.com/brolnickij/yt-dbl/commit/16c815dc4a0ccd08a3f7bd7c14e5179457342e84))
+
+- Extract language maps, lazy imports, early validation, remove dead code
+  ([`a193791`](https://github.com/brolnickij/yt-dbl/commit/a19379117a5b62f09fa5d122979006e86045f621))
+
+- Extract tool configs into standalone files
+  ([`cd98bf8`](https://github.com/brolnickij/yt-dbl/commit/cd98bf818a2d910335bfa9119cf95bb1cc9dac8d))
+
+- Fix type safety in assemble, promote private registry API to public
+  ([`e9c7c68`](https://github.com/brolnickij/yt-dbl/commit/e9c7c68a208b3057fc388e92da41f81374569553))
+
+- Inject ffmpeg path via set_ffmpeg_path instead of importing settings singleton
+  ([`eba5647`](https://github.com/brolnickij/yt-dbl/commit/eba564712f7193d6a076f29af473cdecc2f2d3bb))
+
+- Introduce unified pipeline exception hierarchy
+  ([`4392987`](https://github.com/brolnickij/yt-dbl/commit/4392987b6ad6045a3cb4183cb527c7fbce1f4c09))
+
+- Remove commented-out code
+  ([`489eeb2`](https://github.com/brolnickij/yt-dbl/commit/489eeb2c946148c0e8a39ec8380ca4b19f73ff95))
+
+- Remove dead code (replace_audio, normalize_loudness, deess, TranscriptionError)
+  ([`e8dbc81`](https://github.com/brolnickij/yt-dbl/commit/e8dbc81ac70eb3e240ac8fce3d8c2dd908cf2e90))
+
+- Remove settings singleton from config module
+  ([`6631adf`](https://github.com/brolnickij/yt-dbl/commit/6631adf66ec9f311b156fb59c817dbb10063bd09))
+
+- Remove unused model_manager singleton
+  ([`3ef4512`](https://github.com/brolnickij/yt-dbl/commit/3ef45129bae837832e5fc6beab4b488bed6f7e85))
+
+### Testing
+
+- Add @pytest.mark.parametrize to data-driven tests
+  ([`b22efba`](https://github.com/brolnickij/yt-dbl/commit/b22efba1e4502e0873b40f2d69b68b098d1f9051))
+
+- Add Arabic, Hindi, Thai language detection coverage
+  ([`0307488`](https://github.com/brolnickij/yt-dbl/commit/0307488bad117df8df8f738f16a053058f110b59))
+
+- Add Claude API error path tests for translate step
+  ([`67a4b38`](https://github.com/brolnickij/yt-dbl/commit/67a4b3898deae6a682bbaff891bfc0db931e8c48))
+
+- Add config auto-detection and RAM tier threshold tests
+  ([`a1a8f70`](https://github.com/brolnickij/yt-dbl/commit/a1a8f70c13683ba7b25ccaae9fecfbdc0a5442bd))
+
+- Add coverage for utils/memory.py GPU cleanup
+  ([`6485a01`](https://github.com/brolnickij/yt-dbl/commit/6485a01169565a7822b2e9dea1094254ef160898))
+
+- Add E2E tests for transcription step (VibeVoice-ASR + ForcedAligner)
+  ([`ee7e449`](https://github.com/brolnickij/yt-dbl/commit/ee7e4495d8a4b85a8fcd5251e496e207807fe165))
+
+Real ASR + alignment on 'Me at the zoo' behind --run-slow: - test_transcribe_produces_segments:
+  validates segments, words, speakers - test_idempotent_rerun_uses_cache: verifies segments.json
+  cache - Update pipeline E2E docstring: transcribe is now real, not stub
+
+- Add early API key validation tests for PipelineRunner
+  ([`8a9a267`](https://github.com/brolnickij/yt-dbl/commit/8a9a2671b29736c3a49fce0ca989d428b7d851bd))
+
+- Add postprocess_segment deess fallback coverage
+  ([`74a6cbb`](https://github.com/brolnickij/yt-dbl/commit/74a6cbbef3017f0f4f03a6305e4f5ab58bf4a838))
+
+- Add SynthesisError test for empty TTS result
+  ([`d589af4`](https://github.com/brolnickij/yt-dbl/commit/d589af4b6941495a77238d034623820594996f28))
+
+- Add tests for logging utilities and suppress_library_noise
+  ([`bc46a73`](https://github.com/brolnickij/yt-dbl/commit/bc46a730b5ee83e3c29eea7388b0ced55e79a2f7))
+
+- Add tests for model registry, manager, memory tracking, and config
+  ([`a4a114b`](https://github.com/brolnickij/yt-dbl/commit/a4a114bb25110c6dbd8092a0b33a94779ede49fa))
+
+- Add TTS_LANG_MAP validation tests
+  ([`e7356ff`](https://github.com/brolnickij/yt-dbl/commit/e7356ff2b7f0ece2912ae793794a7a609f8bbc62))
+
+- Isolate Settings from .env and YT_DBL_ env vars in tests
+  ([`c9e52f0`](https://github.com/brolnickij/yt-dbl/commit/c9e52f0de8f5fe26296a137a917c36aaed7c7e4b))
+
+- Make ffmpeg arg assertions order-agnostic
+  ([`d749e50`](https://github.com/brolnickij/yt-dbl/commit/d749e50d8f59432402949521f0d8ec125dd279a2))
+
+- Mock model download check in CLI tests and add API key to settings
+  ([`4610444`](https://github.com/brolnickij/yt-dbl/commit/4610444a087451ab237b5bc37a02989f9c358b32))
+
+- Mock settings in TestStatusCommand
+  ([`fc88bde`](https://github.com/brolnickij/yt-dbl/commit/fc88bdef107e3fd6689016a87100fcc137c390bb))
+
+- Parametrize language detection and format size tests
+  ([`3879d38`](https://github.com/brolnickij/yt-dbl/commit/3879d38bd0dd0cf835284cc92d6b04d1d80d1a0d))
+
+- Remove unused mock in TestStatusCommand
+  ([`e1dea19`](https://github.com/brolnickij/yt-dbl/commit/e1dea192dce84c5607ba46a189e81d5a4d813150))
+
+- Reorganize into unit/integration/e2e pyramid
+  ([`44626c3`](https://github.com/brolnickij/yt-dbl/commit/44626c3af3a98bb0564a20779cc7fe159601b3a7))
+
+- Update version assertion to match semantic versioning format
+  ([`eb2e816`](https://github.com/brolnickij/yt-dbl/commit/eb2e816058d7d87c27305bd436773952a7542f5d))
+
+- Update version assertion to use search for regex matching
+  ([`4c5d30b`](https://github.com/brolnickij/yt-dbl/commit/4c5d30bc6944b013099012b37d62a623eed18970))
+
+- **integration**: Update runner mocks for source_language and TTS
+  ([`1985699`](https://github.com/brolnickij/yt-dbl/commit/1985699b0ee6f57f7dabf8c17b5a3f3bca691d57))
+
+### Breaking Changes
+
+- First stable release of yt-dbl