libspeech 0.2.0__tar.gz → 0.3.0__tar.gz

libspeech-0.3.0/.github/workflows/cmake.yml

@@ -0,0 +1,143 @@
+name: cmake stuff
+
+on:
+  workflow_dispatch:
+  workflow_call:
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build-and-test:
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest, windows-latest]
+
+    runs-on: ${{ matrix.os }}
+
+    steps:
+    - name: Checkout repository
+      uses: actions/checkout@v6
+      with:
+        submodules: true
+
+    - name: Set up Python
+      uses: actions/setup-python@v6
+      with:
+        python-version: "3.11"
+        cache: "pip"
+
+    # speech::models (BUILD_MODELS=ON, the default) needs httpp on the
+    # CMAKE_PREFIX_PATH via find_package(httpp) -- see the "Model
+    # downloading" comment block in the top-level CMakeLists.txt. This has
+    # to happen before the configure step below, since CMake shells out to
+    # `python -c "import httpp; ..."` at configure time to locate it.
+    - name: Install httpp
+      run: pip install httpp
+
+    - name: Build Project
+      uses: threeal/cmake-action@v2.1.0
+      with:
+        build-dir: build
+        options: |
+          CMAKE_BUILD_TYPE=Release
+          BUILD_MODELS=ON
+          BUILD_TESTS=ON
+          BUILD_PYTHON=OFF
+
+    - name: Build
+      run: cmake --build build --config Release
+
+    - name: Run tests
+      run: ctest --test-dir build --output-on-failure --build-config Release
+
+  build-docs:
+    needs: build-and-test
+    runs-on: ubuntu-latest
+
+    steps:
+    - name: Checkout repository
+      uses: actions/checkout@v6
+
+    - name: Install Doxygen + graphviz
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y doxygen graphviz
+
+    - name: Configure CMake
+      run: |
+        cmake -B build \
+          -DBUILD_MODELS=OFF \
+          -DBUILD_TESTS=OFF \
+          -DBUILD_DOCS=ON
+
+    - name: Build documentation
+      run: cmake --build build --target speech_docs
+
+    - name: Upload documentation artifact
+      uses: actions/upload-artifact@v7
+      with:
+        name: cmake-documentation
+        path: build/docs/out/html
+        retention-days: 5
+
+  build-coverage:
+    needs: build-and-test
+    runs-on: ubuntu-latest
+
+    steps:
+    - name: Checkout repository
+      uses: actions/checkout@v6
+      with:
+        submodules: true
+
+    - name: Set up Python
+      uses: actions/setup-python@v6
+      with:
+        python-version: "3.11"
+        cache: "pip"
+
+    # httpp: needed at CMake-configure time (see build-and-test above).
+    # coverage: drives speech_coverage_python. The rest of the Python side
+    # (nanobind, pytest, ...) is installed by CMakeLists.txt itself when
+    # BUILD_PYTHON=ON.
+    - name: Install python dependencies
+      run: pip install httpp coverage
+
+    - name: Install lcov
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y lcov
+
+    # Same shape as ctoon's coverage job: every component (speech::dsp,
+    # speech::models C++ tests + the Python package's tests) is covered
+    # individually, then merged into coverage/total/ + a dashboard at
+    # coverage/index.html. BUILD_COVERAGE is an explicit opt-in (see the
+    # comment in tests/CMakeLists.txt for why it isn't implied by lcov
+    # merely being installed).
+    - name: Configure CMake with coverage
+      run: |
+        cmake -B build \
+          -DBUILD_MODELS=ON \
+          -DBUILD_TESTS=ON \
+          -DBUILD_PYTHON=ON \
+          -DBUILD_COVERAGE=ON
+
+    - name: Generate coverage report
+      run: cmake --build build --target speech_coverage
+
+    - name: Upload coverage to Codecov
+      uses: codecov/codecov-action@v5
+      with:
+        files: build/coverage/total/coverage.lcov
+        flags: unittests
+        name: codecov-libspeech
+        fail_ci_if_error: false
+        verbose: true
+
+    - name: Upload coverage artifact
+      uses: actions/upload-artifact@v7
+      with:
+        name: cmake-coverage-report
+        path: build/coverage/
+        retention-days: 5

{libspeech-0.2.0 → libspeech-0.3.0}/.gitmodules

@@ -3,8 +3,3 @@
 	url = https://github.com/mackron/miniaudio.git
 	shallow = true
 
-[submodule "src/third_party/dr_libs"]
-	path = src/third_party/dr_libs
-	url = https://github.com/mackron/dr_libs.git
-	shallow = true
-

{libspeech-0.2.0 → libspeech-0.3.0}/CMakeLists.txt

@@ -3,6 +3,8 @@ cmake_minimum_required(VERSION 3.19)
 option(BUILD_PYTHON "export python module" off)
 option(BUILD_TESTS "build unit tests (speech::dsp only, no ONNXRuntime/httpp needed)" on)
 option(BUILD_MODELS "build speech::models + the speech library (needs ONNXRuntime download + the httpp package); off configures speech::dsp + tests only" on)
+option(BUILD_DOCS "build the speech_docs Doxygen target (docs/)" off)
+option(BUILD_COVERAGE "instrument speech_dsp with --coverage and enable the speech_coverage_cpp_dsp target (needs lcov/genhtml); OFF by default so a normal build isn't broken just because lcov happens to be installed" off)
 
 set(CMAKE_C_STANDARD 99 CACHE STRING "C version selection")
 set(CMAKE_C_STANDARD_REQUIRED ON)
@@ -108,8 +110,15 @@ if(BUILD_TESTS)
     add_subdirectory(tests)
 endif()
 
+if(BUILD_DOCS)
+    add_subdirectory(docs)
+endif()
+
 if(NOT BUILD_MODELS)
     message(STATUS "BUILD_MODELS=off: configuring speech::dsp + tests only, skipping ONNXRuntime/models/audio I/O.")
+    if(BUILD_TESTS)
+        speech_coverage_finalize()
+    endif()
     return()
 endif()
 
@@ -205,16 +214,18 @@ endif()
 
 # ---------------------------------------------------------------------------
 # speech::io -- audio file I/O (the Audio class: load/save/play/resample/
-# to_mono), backed by miniaudio + dr_libs. Mirrors speech::dsp/speech::models:
-# its own static library, its own namespace (speech::io in the C++ code,
-# matching this CMake alias).
+# to_mono), backed by miniaudio alone (its ma_decoder/ma_encoder embed their
+# own renamed copy of dr_wav/dr_mp3/dr_flac internally -- see the comment at
+# the top of src/audio.cpp -- so the separate dr_libs submodule this used to
+# also need is gone). Mirrors speech::dsp/speech::models: its own static
+# library, its own namespace (speech::io in the C++ code, matching this
+# CMake alias).
 # ---------------------------------------------------------------------------
 add_library(speech_io STATIC src/audio.cpp)
 add_library(speech::io ALIAS speech_io)
 
 target_include_directories(speech_io
         PUBLIC ${PROJECT_SOURCE_DIR}/include
-        PRIVATE src/third_party/dr_libs
         PRIVATE src/third_party/aixlog/include
 )
 target_link_libraries(speech_io
@@ -481,7 +492,7 @@ if(DEFINED PYTHON_PROJECT_NAME)
                 LIBRARY DESTINATION ${PYTHON_PROJECT_NAME}
                 RUNTIME DESTINATION ${PYTHON_PROJECT_NAME})
 
-        # The pure-Python files (__init__.py, __main__.py, core.py) need
+        # The pure-Python files (currently just __init__.py) need
         # their own explicit install rule -- scikit-build-core's automatic
         # Python-package discovery only looks for an immediate src/<name>
         # layout, so now that the package lives nested under
@@ -591,6 +602,13 @@ if(BUILD_TESTS AND DEFINED PYTHON_PROJECT_NAME AND NOT DEFINED SKBUILD)
     add_subdirectory(tests/python)
 endif()
 
+# Merge every coverage component configured above into speech_coverage_total
+# + the dashboard (no-op unless -DBUILD_COVERAGE=ON). The BUILD_MODELS=OFF
+# early return above calls this too, since it never reaches this point.
+if(BUILD_TESTS)
+    speech_coverage_finalize()
+endif()
+
 if(DEFINED SKBUILD)
     RETURN()
 endif()

{libspeech-0.2.0 → libspeech-0.3.0}/CONTRIBUTING.md

@@ -12,16 +12,15 @@ git clone https://github.com/MohammadRaziei/libspeech.git
 cd libspeech
 
 # Small, lightweight submodules:
-git submodule update --init src/third_party/miniaudio src/third_party/dr_libs src/third_party/indicators
+git submodule update --init src/third_party/miniaudio
 
-# Mbed TLS (pinned to v3.6.2, kept shallow -- see .gitmodules) has its own
-# nested submodule ("framework") that also needs initializing:
-git submodule update --init src/third_party/mbedtls
-git submodule update --init --depth 1 src/third_party/mbedtls/framework
+# speech::models (BUILD_MODELS=ON, the default) needs httpp on
+# CMAKE_PREFIX_PATH for model downloads (HTTPS client + progress bar):
+pip install httpp
 ```
 
-A few entries under `src/third_party/` (AudioFlux subset, aixlog, httplib)
-are **not** submodules -- those are copied directly into the repo, alongside
+A few entries under `src/third_party/` (AudioFlux subset, aixlog) are
+**not** submodules -- those are copied directly into the repo, alongside
 the actual submodules. See `src/third_party/README.md` for why and which
 ones.
 
@@ -103,9 +102,12 @@ what's already vendored under `src/third_party/`:
    what/why/before-after -- in `audioflux_issues.md` (or a new file
    following that structure for a different vendored project), so it can
    become an upstream PR later.
-3. Security-critical code (crypto, TLS) is the exception: keep those as
-   full git submodules (see Mbed TLS above), not cherry-picked/vendored,
-   so security patches stay easy to pull in.
+3. Security/network-critical code is a different story: `httpp`
+   (HTTPS client + progress bar for model downloads) is a genuine,
+   separately-`pip install`able runtime dependency, not something to vendor
+   or patch in-tree -- see the `find_package(httpp)` comment block in
+   `CMakeLists.txt` for why (this replaced an earlier vendored
+   cpp-httplib + Mbed TLS setup for exactly this reason).
 
 ## Style notes
 
@@ -136,6 +138,46 @@ python version.py minor +            # bump the minor version
 python version.py tag create patch   # bump patch, commit, and tag a release
 ```
 
+## Docs and coverage
+
+`-DBUILD_DOCS=ON` gives you `speech_docs` (Doxygen, warnings-free -- keep it
+that way when you touch a public header). `-DBUILD_COVERAGE=ON` (needs
+`lcov`; add `-DBUILD_PYTHON=ON` and `pip install coverage` for the Python
+side) gives you `speech_coverage`, which runs every component's tests
+instrumented, merges them, and writes a dashboard to
+`build/coverage/index.html`. It mirrors ctoon's setup -- see the "Docs &
+coverage" table in `README.md`. A new C++ module with its own test target
+gets a coverage target by calling `speech_enable_coverage(<lib>)` and
+`speech_add_cpp_coverage_target(...)` next to its tests (see
+`tests/models/CMakeLists.txt` for a three-line example).
+
+Two lcov 2.0 quirks worth knowing before you debug them yourself: its
+`--list` output is wrong (use `--summary`, as the targets do), and an
+`--extract`/`--exclude` pattern that matches nothing is a hard error unless
+you pass `--ignore-errors unused`.
+
+## Packaging / CI changes
+
+If you touch `CMakeLists.txt`'s install rules, `pyproject.toml`, or
+`.github/workflows/wheels.yml`, actually build a wheel locally and inspect
+its *contents* before opening a PR -- don't just read the config and
+assume it's right:
+
+```bash
+python -m build --wheel -o /tmp/wheelcheck
+python -c "import zipfile; z = zipfile.ZipFile('/tmp/wheelcheck/<name>.whl'); [print(i.filename, i.file_size) for i in z.infolist()]"
+```
+
+This project has twice shipped a packaging bug invisible from reading the
+config alone -- a duplicated ~20MB ONNX Runtime binary (symlinks get
+dereferenced both by `install(FILES ...)` and by the wheel-zip step), and a
+wheel silently missing its intended `abi3` tag (`CIBW_ENVIRONMENT_CP312`
+looked like a real cibuildwheel option by analogy with
+`CIBW_ENVIRONMENT_LINUX`, but isn't -- cibuildwheel only recognizes
+platform suffixes there, not Python-version ones, and silently no-ops
+instead of erroring). Both only showed up by actually opening the built
+artifact.
+
 ## Current project status
 
 See [`checklist.md`](checklist.md) for the up-to-date, detailed state of

{libspeech-0.2.0 → libspeech-0.3.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: libspeech
-Version: 0.2.0
+Version: 0.3.0
 Summary: Best library for speech
 Author-Email: Mohammad Raziei <mohammadraziei1375@gmail.com>
 Classifier: Development Status :: 4 - Beta
@@ -43,7 +43,7 @@ Description-Content-Type: text/markdown
 [![Python Versions](https://img.shields.io/pypi/pyversions/libspeech)](https://pypi.org/project/libspeech/)
 [![C++ Standard](https://img.shields.io/badge/C++-17-blue)](https://en.cppreference.com/w/cpp/17)
 
-[![Build Status](https://github.com/mohammadraziei/libspeech/actions/workflows/build.yml/badge.svg)](https://github.com/mohammadraziei/libspeech/actions)
+[![Build Status](https://github.com/mohammadraziei/libspeech/actions/workflows/cmake.yml/badge.svg)](https://github.com/mohammadraziei/libspeech/actions)
 [![Code Quality](https://sonarcloud.io/api/project_badges/measure?project=MohammadRaziei_libspeech&metric=alert_status)](https://sonarcloud.io/summary/new_code?id=MohammadRaziei_libspeech)
 [![CodeFactor](https://www.codefactor.io/repository/github/mohammadraziei/libspeech/badge/master)](https://www.codefactor.io/repository/github/mohammadraziei/libspeech/overview/master)
 [![snyk.io](https://snyk.io/advisor/python/libspeech/badge.svg)](https://snyk.io/advisor/python/libspeech)
@@ -59,7 +59,7 @@ Description-Content-Type: text/markdown
 - 🔊 **Speech models** (`speech::models`): a `Denoiser` interface (Facebook/SpeechBrain backends) and Silero VAD
 - 🖥️ **Cross-platform**: Windows, Linux, macOS
 - 🐍 **Python bindings** (nanobind) mirroring the C++ API 1:1
-- 📦 **Zero system dependencies**: no `apt install libcurl-dev` / OpenSSL needed -- HTTPS model downloads go through a vendored [cpp-httplib](https://github.com/yhirose/cpp-httplib) + [Mbed TLS](https://github.com/Mbed-TLS/mbedtls) built from source
+- 📦 **Zero system dependencies**: no `apt install libcurl-dev` / OpenSSL needed -- HTTPS model downloads go through [`httpp`](https://github.com/mohammadraziei/httpp) (`pip install httpp`), a separately-installed package with its own prebuilt HTTP(S) client and progress bar
 
 ## 📦 Installation
 
@@ -72,11 +72,12 @@ pip install libspeech
 ```bash
 git clone https://github.com/MohammadRaziei/libspeech.git
 cd libspeech
-git submodule update --init src/third_party/miniaudio src/third_party/dr_libs src/third_party/indicators
+git submodule update --init src/third_party/miniaudio
 
-# Mbed TLS is pinned to v3.6.2 and kept as a shallow submodule (see .gitmodules):
-git submodule update --init src/third_party/mbedtls
-git submodule update --init --depth 1 src/third_party/mbedtls/framework
+# speech::models (BUILD_MODELS=ON, the default) needs httpp on
+# CMAKE_PREFIX_PATH for model downloads (HTTPS client + progress bar) --
+# no more Mbed TLS/libcurl to build, just:
+pip install httpp
 
 mkdir build && cd build
 cmake ..                      # downloads ONNX Runtime automatically on first configure
@@ -159,6 +160,34 @@ speech_test                       # everything
 Run everything with `cmake --build build --target speech_test`, or just
 `ctest` for the same suites without the extra build-tool chatter.
 
+### Docs & coverage
+
+```bash
+cmake .. -DBUILD_DOCS=ON && cmake --build . --target speech_docs   # Doxygen HTML -> build/docs/out/html
+```
+
+Coverage follows the same layout as [ctoon](https://github.com/mohammadraziei/ctoon):
+one target per component, a merged total, and a dashboard tying the reports
+together.
+
+```bash
+cmake .. -DBUILD_PYTHON=ON -DBUILD_COVERAGE=ON   # + `pip install coverage`, lcov/genhtml on PATH
+cmake --build . --target speech_coverage         # everything below, then the merge + dashboard
+```
+
+| Target | Covers | Output (under `build/coverage/`) |
+|---|---|---|
+| `speech_coverage_cpp_dsp` | `speech::dsp` (utest.h suite) | `cpp_dsp/` |
+| `speech_coverage_cpp_models` | `speech::models` + download utils | `cpp_models/` |
+| `speech_coverage_python` | the Python package's `.py` files (coverage.py) | `python/` |
+| `speech_coverage_total` | lcov merge of all of the above | `total/` |
+
+Open `build/coverage/index.html` for the dashboard. Each component (and the
+total) also leaves a plain `coverage.lcov` next to its HTML, which is what CI
+uploads to Codecov. `BUILD_COVERAGE` is off by default: it instruments
+`speech_dsp`/`speech_models` with `--coverage`, which you don't want in a
+normal build. Without `BUILD_PYTHON` you just get the C++ components.
+
 ## 🔧 Architecture
 
 libspeech is split into three independent C++ libraries (each with its own
@@ -168,7 +197,7 @@ library:
 | Namespace | CMake target | Python module | Depends on |
 |---|---|---|---|
 | `speech::dsp` | `speech_dsp` | `speech_dsp_py` | nothing but a vendored subset of [AudioFlux](https://github.com/libAudioFlux/audioflux)'s C sources |
-| `speech::io` | `speech_io` | `speech_io_py` | `speech::dsp` (for resampling), miniaudio, dr_libs |
+| `speech::io` | `speech_io` | `speech_io_py` | `speech::dsp` (for resampling), miniaudio |
 | `speech::models` | `speech_models` | `speech_models_py` | ONNX Runtime, httplib+Mbed TLS (for downloading model weights) |
 
 See [`checklist.md`](checklist.md) for the detailed, up-to-date state of

{libspeech-0.2.0 → libspeech-0.3.0}/README.md

@@ -17,7 +17,7 @@
 [![Python Versions](https://img.shields.io/pypi/pyversions/libspeech)](https://pypi.org/project/libspeech/)
 [![C++ Standard](https://img.shields.io/badge/C++-17-blue)](https://en.cppreference.com/w/cpp/17)
 
-[![Build Status](https://github.com/mohammadraziei/libspeech/actions/workflows/build.yml/badge.svg)](https://github.com/mohammadraziei/libspeech/actions)
+[![Build Status](https://github.com/mohammadraziei/libspeech/actions/workflows/cmake.yml/badge.svg)](https://github.com/mohammadraziei/libspeech/actions)
 [![Code Quality](https://sonarcloud.io/api/project_badges/measure?project=MohammadRaziei_libspeech&metric=alert_status)](https://sonarcloud.io/summary/new_code?id=MohammadRaziei_libspeech)
 [![CodeFactor](https://www.codefactor.io/repository/github/mohammadraziei/libspeech/badge/master)](https://www.codefactor.io/repository/github/mohammadraziei/libspeech/overview/master)
 [![snyk.io](https://snyk.io/advisor/python/libspeech/badge.svg)](https://snyk.io/advisor/python/libspeech)
@@ -33,7 +33,7 @@
 - 🔊 **Speech models** (`speech::models`): a `Denoiser` interface (Facebook/SpeechBrain backends) and Silero VAD
 - 🖥️ **Cross-platform**: Windows, Linux, macOS
 - 🐍 **Python bindings** (nanobind) mirroring the C++ API 1:1
-- 📦 **Zero system dependencies**: no `apt install libcurl-dev` / OpenSSL needed -- HTTPS model downloads go through a vendored [cpp-httplib](https://github.com/yhirose/cpp-httplib) + [Mbed TLS](https://github.com/Mbed-TLS/mbedtls) built from source
+- 📦 **Zero system dependencies**: no `apt install libcurl-dev` / OpenSSL needed -- HTTPS model downloads go through [`httpp`](https://github.com/mohammadraziei/httpp) (`pip install httpp`), a separately-installed package with its own prebuilt HTTP(S) client and progress bar
 
 ## 📦 Installation
 
@@ -46,11 +46,12 @@ pip install libspeech
 ```bash
 git clone https://github.com/MohammadRaziei/libspeech.git
 cd libspeech
-git submodule update --init src/third_party/miniaudio src/third_party/dr_libs src/third_party/indicators
+git submodule update --init src/third_party/miniaudio
 
-# Mbed TLS is pinned to v3.6.2 and kept as a shallow submodule (see .gitmodules):
-git submodule update --init src/third_party/mbedtls
-git submodule update --init --depth 1 src/third_party/mbedtls/framework
+# speech::models (BUILD_MODELS=ON, the default) needs httpp on
+# CMAKE_PREFIX_PATH for model downloads (HTTPS client + progress bar) --
+# no more Mbed TLS/libcurl to build, just:
+pip install httpp
 
 mkdir build && cd build
 cmake ..                      # downloads ONNX Runtime automatically on first configure
@@ -133,6 +134,34 @@ speech_test                       # everything
 Run everything with `cmake --build build --target speech_test`, or just
 `ctest` for the same suites without the extra build-tool chatter.
 
+### Docs & coverage
+
+```bash
+cmake .. -DBUILD_DOCS=ON && cmake --build . --target speech_docs   # Doxygen HTML -> build/docs/out/html
+```
+
+Coverage follows the same layout as [ctoon](https://github.com/mohammadraziei/ctoon):
+one target per component, a merged total, and a dashboard tying the reports
+together.
+
+```bash
+cmake .. -DBUILD_PYTHON=ON -DBUILD_COVERAGE=ON   # + `pip install coverage`, lcov/genhtml on PATH
+cmake --build . --target speech_coverage         # everything below, then the merge + dashboard
+```
+
+| Target | Covers | Output (under `build/coverage/`) |
+|---|---|---|
+| `speech_coverage_cpp_dsp` | `speech::dsp` (utest.h suite) | `cpp_dsp/` |
+| `speech_coverage_cpp_models` | `speech::models` + download utils | `cpp_models/` |
+| `speech_coverage_python` | the Python package's `.py` files (coverage.py) | `python/` |
+| `speech_coverage_total` | lcov merge of all of the above | `total/` |
+
+Open `build/coverage/index.html` for the dashboard. Each component (and the
+total) also leaves a plain `coverage.lcov` next to its HTML, which is what CI
+uploads to Codecov. `BUILD_COVERAGE` is off by default: it instruments
+`speech_dsp`/`speech_models` with `--coverage`, which you don't want in a
+normal build. Without `BUILD_PYTHON` you just get the C++ components.
+
 ## 🔧 Architecture
 
 libspeech is split into three independent C++ libraries (each with its own
@@ -142,7 +171,7 @@ library:
 | Namespace | CMake target | Python module | Depends on |
 |---|---|---|---|
 | `speech::dsp` | `speech_dsp` | `speech_dsp_py` | nothing but a vendored subset of [AudioFlux](https://github.com/libAudioFlux/audioflux)'s C sources |
-| `speech::io` | `speech_io` | `speech_io_py` | `speech::dsp` (for resampling), miniaudio, dr_libs |
+| `speech::io` | `speech_io` | `speech_io_py` | `speech::dsp` (for resampling), miniaudio |
 | `speech::models` | `speech_models` | `speech_models_py` | ONNX Runtime, httplib+Mbed TLS (for downloading model weights) |
 
 See [`checklist.md`](checklist.md) for the detailed, up-to-date state of

{libspeech-0.2.0 → libspeech-0.3.0}/checklist.md

@@ -20,7 +20,14 @@ its tests are green (see `AGENTS`/conversation ground rules).
 - [x] Establish logging convention: `aixlog`, `DEBUG` for lifecycle/results/errors, `TRACE` only where it exposes internal computation worth debugging (not on every call)
 - [x] Establish AudioFlux vendoring workflow: copy only the `.c`/`.h` files a ported operator needs into `src/third_party/audioflux/`, patch in place if needed, log every patch in `/audioflux_issues.md`
 - [x] Remove `src/third_party/audioflux` submodule entirely (nothing in the build references it anymore — everything comes from `src/third_party/audioflux`)
-- [ ] Decide fate of `src/third_party/indicators`, `dr_libs`, `miniaudio` (still used by `Audio`/CLI progress bars — keep for now, revisit)
+- [x] Decide fate of `src/third_party/indicators`, `dr_libs`, `miniaudio`:
+  `indicators` fully removed — `src/audio.cpp`'s progress bar now uses
+  `httpp::progress::bar` (`<httpp/progress.hpp>`) from the `httpp` package
+  itself, same as the download-progress bar in
+  `speech::utils::downloadFile` (`.enable_progress()`). `dr_libs`/
+  `miniaudio` stay: they're audio file I/O (decode/encode/playback), an
+  unrelated concern from progress bars, still the only thing providing
+  that.
 
 > **Note for future DSP operators:** with the full AudioFlux submodule gone,
 > the next time we need to vendor a new `.c`/`.h` file (e.g. for STFT/MFCC),
@@ -117,18 +124,48 @@ its tests are green (see `AGENTS`/conversation ground rules).
   system dependencies, only source builds.
 - [ ] Verify full `speech` library CMake build end-to-end on a machine with normal internet access (this sandbox's own egress restrictions, now unrelated to CURL, are the only remaining blocker here)
 - [ ] `pip install` end-to-end smoke test (core promise of the project: zero system deps)
-- [ ] CI update: drop googletest/aixlog submodule steps, add utest.h-based test target
+- [x] CI update: the googletest/aixlog-submodule cleanup itself turned out
+  to already be done (confirmed: no gtest/aixlog steps anywhere in
+  `.github/workflows/`, `.gitmodules` only lists `miniaudio`/`dr_libs`, and
+  `tests/CMakeLists.txt` already has a real `utest.h`-based `test_cpp_dsp`
+  target) -- this item was stale. What was actually still missing,
+  compared to `mohammadraziei/ctoon`'s CI structure: a `build-docs` job
+  (Doxygen) and a `build-coverage` job (lcov), both added to
+  `.github/workflows/cmake.yml` this session and verified by actually
+  building both locally first (see the Doxygen and Coverage entries
+  elsewhere in this file for what each surfaced). Deliberately not a full
+  copy of ctoon's job matrix -- no Rust/Go/Julia toolchain setup steps,
+  since libspeech doesn't have any of those bindings.
 
 ## Documentation
 
 - [x] `/audioflux_issues.md` — running log of every AudioFlux bug/quirk found + fixed, PR-ready
 - [x] `src/third_party/audioflux/README.md` — explains why vendored instead of submoduled
-- [ ] Update root `README.md` Quick Start to match actual current API (it currently references classes/methods that don't exist yet — `AudioProcessor`, `extract_features`, etc.)
-- [ ] `CONTRIBUTING.md` (referenced by README but missing)
-- [ ] Doxygen-style comments on public headers (`Audio`, `BaseModel`, ...)
-
-## Python bindings (nanobind)
-
+- [x] Update root `README.md` Quick Start to match actual current API —
+  the Quick Start itself was already fixed (real `Audio`/`MFCC`/`Denoiser`/
+  `SileroVad` API, not the old `AudioProcessor`/`extract_features`), but
+  the separate "From source" build steps and the "Zero system
+  dependencies" bullet were still stale (Mbed TLS/`indicators` submodule
+  init steps that no longer exist since the `httpp` migration) — fixed,
+  plus a broken CI badge link (`workflows/build.yml` doesn't exist;
+  pointed it at `workflows/cmake.yml`).
+- [x] `CONTRIBUTING.md` (referenced by README) — was already written
+  earlier (see below), just stale by the time of this check (still
+  mentioned Mbed TLS's nested submodule and `src/third_party/indicators`,
+  both removed since); refreshed to match current reality.
+- [x] Doxygen-style comments on public headers: the `speech_docs` target
+  (+ `build-docs` CI job) surfaced 13 concrete warnings, all now fixed and
+  verified -- `speech_docs` builds with **0** warnings. Real gaps closed:
+  missing/wrong `@param` docs on `FacebookDenoiser`/`SpeechBrainDenoiser`
+  constructors (documented `base_dir` for a parameter that doesn't exist,
+  omitted `sample_rate`), a stale `@param sample_rate` on both `process()`
+  methods (it takes only `input_audio`), the undocumented `sample_rate` on
+  `ONNXModel`'s constructor, and a copy-pasted "Destructor for
+  FacebookDenoiser" on `SpeechBrainDenoiser`. Also: README's relative links
+  to `checklist.md`/`audioflux_issues.md`/`CONTRIBUTING.md` only resolve if
+  those files are in Doxygen's `INPUT` (added), and Doxygen's markdown
+  parser choked on two multi-line `**bold**` spans and a `__repr__`-style
+  dunder name in `checklist.md` (reworded).
 - [x] Fixed `bind_audio` (the `_audio` module): it referenced a nonexistent
   `speech::Audio::sampleRate()` (the real method is `sample_rate()`) --
   this binding never actually compiled before.
@@ -219,9 +256,61 @@ ctoon's `tests/python/CMakeLists.txt` pattern:
   all 37 DSP tests *and* all 8 Python tests in one command; `ctest` shows
   both `speech_dsp` and `speech_python` entries; `-DBUILD_MODELS=OFF`
   (DSP-only) still works correctly with no Python target present.
-- [ ] Coverage reporting (lcov/coverage.py + merged dashboard) -- ctoon has
-  an elaborate version of this; not set up here yet, out of scope for now.
-
+- [x] Coverage reporting, modeled on ctoon's flow: a `speech_coverage_<component>`
+  target per component (`cpp_dsp`, `cpp_models`, `python`), a merged
+  `speech_coverage_total` (lcov `--add-tracefile`), a JSON-driven dashboard at
+  `build/coverage/index.html` (template, iframe wrapper, `genhtml.css` and the
+  `FixLCovPaths`/`GenerateIframe` scripts copied from ctoon and rebranded), a
+  `speech_coverage` umbrella, and a `build-coverage` CI job that runs the
+  umbrella and uploads `coverage/total/coverage.lcov` to Codecov. Verified by
+  actually running the whole umbrella locally (`BUILD_MODELS=ON BUILD_TESTS=ON
+  BUILD_PYTHON=ON BUILD_COVERAGE=ON`): all three components + merge + dashboard
+  build, dashboard data parses and lists total/cpp_dsp/cpp_models/py, and the
+  DSP-only configuration (`BUILD_MODELS=OFF`) and a normal coverage-off build
+  (5/5 ctest) still work. Numbers at the time: `speech::dsp` 82.7% lines /
+  89.7% functions / 38.4% branches, `speech::models` 21.3% / 22.6% / 25.5%
+  (only 7 C++ tests so far, and the denoiser/VAD paths need model weights),
+  Python package 37% (see the dead-code item below).
+  Differences from ctoon, all deliberate: (1) gated on an explicit
+  `BUILD_COVERAGE` option, not merely on lcov being installed -- `speech_dsp`/
+  `speech_models` are real STATIC libraries with other consumers (`example`,
+  `speech`, the bindings), and gating on `find_program(lcov)` alone broke the
+  *normal* build wherever lcov happened to be installed (`undefined reference
+  to '__gcov_init'`); the instrumented libraries now export `--coverage` as a
+  PUBLIC *link* option so consumers always link libgcov; (2) `lcov --summary`
+  instead of `--list`, because `--list` in lcov 2.0-1 (Ubuntu 24.04) prints
+  wrong columns (4.8% for a file whose record says LF:21 LH:21, function
+  "rates" of 7200%) while `--summary` matches the raw LF/LH/FNF/FNH counters;
+  (3) `--exclude '*/src/third_party/*' --ignore-errors unused` at capture time
+  (a vendored aixlog lambda triggers a gcov "mismatched end line" hard error
+  otherwise; and lcov 2.x treats a pattern that matches nothing as an error),
+  and `--rc branch_coverage=1`/`--branch-coverage` on *every* lcov/genhtml
+  call -- that missing flag on `--extract` is why branch coverage earlier
+  showed "no data found". No Go/Rust/Julia/MATLAB/Zig parts, since libspeech
+  has none of those bindings.
+- [x] Removed the `dr_libs` submodule -- it was pure duplication.
+  `miniaudio.h` already vendors its own renamed copy of dr_wav/dr_mp3/dr_flac
+  internally (search `dr_wav_h begin`/`dr_flac_h begin`/`dr_mp3_h begin` in
+  it -- same upstream author, kept in sync) and exposes all three uniformly
+  through `ma_decoder`/`ma_encoder` (`ma_encoding_format_{wav,flac,mp3}`),
+  including WAV encoding. Rewrote `src/audio.cpp`'s `loadWAV`/`loadMP3`/
+  `loadFLAC` into one shared `loadWithMiniaudio()` helper (encoding format
+  passed explicitly, so a mismatched extension still fails the same way the
+  old per-format `dr_*_init_*` calls did) and `saveWAV` to use
+  `ma_encoder_init_file`/`ma_encoder_write_pcm_frames`; dropped the
+  `dr_wav.h`/`dr_mp3.h`/`dr_flac.h` includes and their `DR_*_IMPLEMENTATION`
+  macros entirely. `speech_io`'s CMake target, `.gitmodules`, and the
+  submodule-init line in `README.md`/`CONTRIBUTING.md`/
+  `src/third_party/README.md` all updated to drop `dr_libs`; actually ran
+  `git submodule deinit`/`git rm` for it, not just deleted the directory.
+  Verified with real audio, not just a compile: `ffmpeg`-generated WAV/MP3/
+  FLAC (mono) + a stereo FLAC loaded correctly (right sample rate/channel
+  count/duration for all four); WAV and FLAC (both lossless) decoded to
+  bit-identical samples; MP3's duration came out ~10% longer than the
+  source, which is the format's own encoder priming/padding delay, not a
+  regression (same file, same padding would occur with dr_mp3 too); the
+  existing `test_save_and_load_wav_round_trip` pytest and the full `ctest`
+  suite (5/5) still pass.
 ## Python bindings for speech::models
 
 - [x] New `bind_models` module (`_models`, re-exported as
@@ -233,7 +322,7 @@ ctoon's `tests/python/CMakeLists.txt` pattern:
 - [x] `SileroVad`: full constructor (all tunable parameters), `.process()`,
   `.get_speech_timestamps()`, `.reset()`.
 - [x] `SpeechTimestamp`: `start`/`end` (samples), `start_s`/`end_s`
-  (seconds), `__repr__`, `__eq__`.
+  (seconds), plus the `repr`/`eq` dunder methods.
 - [x] Found + fixed a real, pre-existing bug while testing this binding:
   `timestamp_t::start_s()`/`end_s()` (`include/libspeech/models/silero_vad.h`)
   performed integer division (`start / sample_rate`, both `int`), always
@@ -490,9 +579,8 @@ Previously: `speech_dsp`/`speech_io`/`speech_models` (C++ static libs) but
   benchmark seemed to show ~24% improvement (82ms -> 62ms), but a more
   careful re-run (20 reps, warm-up excluded) showed the before/after times
   are statistically indistinguishable (62.84ms vs. 61.78ms) -- the initial
-  result was cold-start/measurement noise, not a real effect. **Needs
-  re-benchmarking on an actual multi-core machine** to confirm the
-  expected real-world speedup before this can be claimed anywhere.
+  result was cold-start/measurement noise, not a real effect. **Needs re-benchmarking on an actual multi-core machine**
+  to confirm the expected real-world speedup before this can be claimed anywhere.
 - Other identified-but-not-yet-pursued speed opportunities (same
   benchmark-first discipline applies before acting on any of these):
   flat/contiguous buffers instead of `vector<vector<float>>` for

libspeech-0.3.0/cmake/FixLCovPaths.cmake

@@ -0,0 +1,9 @@
+# Read the content of the coverage file
+file(READ ${COV_FILE} CONTENT)
+
+# Replace the absolute source path with an empty string
+# To relative paths like: include/...
+string(REPLACE "SF:${SOURCE_PATH}" "SF:${PREFIX_PATH}" CONTENT "${CONTENT}")
+
+# Write the modified content back to the file
+file(WRITE ${COV_FILE} "${CONTENT}")