Merge pull request #94 from AIWintermuteAI/updating-to-fix-CI

Fix the CI builds

Author: AIWintermuteAI

.github/workflows/ci.yml

@@ -7,7 +7,7 @@ on:
     branches:
       - develop
   schedule:
-    - cron: '00 12 * * 1' # Monday 12 PM UTC
+    - cron: '00 12 1 * *' # First day of each month 12 PM UTC
 concurrency:
   group: ci-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
@@ -23,9 +23,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
-        exclude:
-          - os: 'windows-latest'
+        os: [ubuntu-latest, macos-latest]
     name: Integration tests (${{ matrix.os }})
     runs-on: ${{ matrix.os }}
     timeout-minutes: 90
@@ -37,7 +35,7 @@ jobs:
       - name: Setup CI
         uses: ./.github/actions/setup-repo
         with:
-          python-version: '3.8' # sync with default version in WORKSPACE.
+          python-version: '3.11.8' # sync with default version in WORKSPACE.
       - name: Install development dependencies and bindings
         run: |
           bazel run pypi_update

.github/workflows/submodule-update.yml

@@ -7,8 +7,8 @@ name: Submodule Updates
 on:
   workflow_dispatch:
   schedule:
-    - cron: '00 12 * * 1' # Monday 12 PM UTC
-    
+    - cron: '00 12 1 * *' # First day of each month 12 PM UTC
+
 ###############
 # Set the Job #
 ###############

.github/workflows/wheels.yml

@@ -7,6 +7,8 @@ on:
   pull_request:
     branches:
       - develop
+  schedule:
+    - cron: '00 12 1 * *' # First day of each month 12 PM UTC
 concurrency:
   group: wheels-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
@@ -51,10 +53,11 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: ['ubuntu-latest', 'macos-latest', 'windows-latest']
-        python-version: ['3.10', '3.11']
+        os: ['ubuntu-latest', 'macos-latest']
+        python-version: ['3.10', '3.11.8']
         exclude:
-          - os: 'windows-latest'
+          - os: 'macos-latest'
+            python-version: '3.10'
     steps:
       - name: Checkout
         uses: actions/checkout@v4

.gitignore

@@ -136,3 +136,4 @@ bun.lockb
 *.bin
 
 src/whispercpp/__about__.py
+.vscode/settings.json

BUILD.bazel

@@ -221,9 +221,9 @@ py_wheel(
     homepage = "https://github.com/aarnphm/whispercpp",
     license = "Apache 2.0",
     platform = select({
-        "//:aarch64-apple-darwin": "macosx_10_9_arm64",
+        "//:aarch64-apple-darwin": "macosx_14_0_arm64",
         "//:aarch64-unknown-linux-gnu": "manylinux2014_aarch64",
-        "//:x86_64-apple-darwin": "macosx_10_9_x86_64",
+        "//:x86_64-apple-darwin": "macosx_14_0_x86_64",
         "//:x86_64-pc-windows-msvc": "win_amd64",
         "//:x86_64-unknown-linux-gnu": "manylinux2014_x86_64",
     }),

examples/BUILD

@@ -16,6 +16,5 @@ py_test(
         "//src/whispercpp:whispercpp_lib",
         requirement("bazel-runfiles"),
         requirement("numpy"),
-        requirement("pydub"),
     ],
 )

examples/bindings/explore.py

@@ -34,7 +34,7 @@ def main(argv: list[str]) -> int:
     assert Path(path).exists()
     params = get_model().params.with_print_realtime(True).build()
     assert _model is not None
-    _model.context.full(params, w.api.load_wav_file(Path(path).__fspath__()).mono)
+    _model.context.full(params, w.api.load_wav_file_mono(Path(path).__fspath__()))
     return 0
 
 

requirements/pypi.txt

@@ -1,4 +1,3 @@
-pydub==0.25.1
 ruff
 black
 bazel-runfiles==0.19.0
@@ -10,4 +9,4 @@ pytest-xdist[psutil]
 pytest-asyncio
 build
 virtualenv
-twine
+twine

rules/deps.bzl

@@ -96,7 +96,7 @@ def internal_deps():
     # Override python rules defined under @com_github_bentoml_plugins
     http_archive(
         name = "rules_python",
-        sha256 = "b0a45e1fdaade7ac4975a9dfa08a170b78e209ba717946f7d397a75dc8a9fda6",
-        strip_prefix = "rules_python-4e3d01c9f6a6cda0fb34b7c2961d6b1ef5313218",
-        url = "https://github.com/bazelbuild/rules_python/archive/4e3d01c9f6a6cda0fb34b7c2961d6b1ef5313218.tar.gz",
+        sha256 = "3d6fe72f1a056b3462f02afba5049210acbaec131087fb19082fa6792198a9fa",
+        strip_prefix = "rules_python-084b877c98b580839ceab2b071b02fc6768f3de6",
+        url = "https://github.com/bazelbuild/rules_python/archive/084b877c98b580839ceab2b071b02fc6768f3de6.tar.gz",
     )

src/whispercpp/__init__.py

@@ -221,7 +221,7 @@ def transcribe_from_file(
             Transcribed text.
         """
         return self.transcribe(
-            api.load_wav_file(filename).mono, num_proc=num_proc, strict=strict
+            api.load_wav_file_mono(filename), num_proc=num_proc, strict=strict
         )
 
     def stream_transcribe(

src/whispercpp/api_cpp2py_export.cc

@@ -9,6 +9,24 @@ WavFileWrapper WavFileWrapper::load_wav_file(const char *filename) {
     return WavFileWrapper(&pcmf32, &pcmf32s);
 }
 
+std::vector<std::vector<float>> load_wav_file_stereo(const char *filename) {
+    std::vector<float> pcmf32;
+    std::vector<std::vector<float>> pcmf32s;
+    if (!::read_wav(filename, pcmf32, pcmf32s, true)) {
+        throw std::runtime_error("Failed to load wav file");
+    }
+    return pcmf32s;
+}
+
+std::vector<float> load_wav_file_mono(const char *filename) {
+    std::vector<float> pcmf32;
+    std::vector<std::vector<float>> pcmf32s;
+    if (!::read_wav(filename, pcmf32, pcmf32s, false)) {
+        throw std::runtime_error("Failed to load wav file");
+    }
+    return pcmf32;
+}
+
 namespace py = pybind11;
 using namespace pybind11::literals;
 
@@ -58,6 +76,12 @@ PYBIND11_MODULE(api_cpp2py_export, m) {
                whisper_sampling_strategy::WHISPER_SAMPLING_BEAM_SEARCH)
         .export_values();
 
+    m.def("load_wav_file_stereo", load_wav_file_stereo, "filename"_a,
+          py::return_value_policy::reference);
+
+    m.def("load_wav_file_mono", load_wav_file_mono, "filename"_a,
+          py::return_value_policy::reference);
+
     m.def("load_wav_file", &WavFileWrapper::load_wav_file, "filename"_a,
           py::return_value_policy::reference);
 

tests/BUILD

@@ -19,7 +19,6 @@ py_test(
         "//src/whispercpp:whispercpp_lib",
         requirement("bazel-runfiles"),
         requirement("numpy"),
-        requirement("pydub"),
     ],
 )
 

tests/export_test.py

@@ -7,7 +7,7 @@
 import pytest as p
 
 import whispercpp as w
-from pydub import AudioSegment as am
+import wave
 
 if t.TYPE_CHECKING:
     import numpy as np
@@ -21,10 +21,14 @@
 
 def preprocess(file: Path, sample_rate: int = 16000) -> NDArray[np.float32]:
 
-    sound = am.from_file(file, format='wav', frame_rate=sample_rate)
-    sound = sound.set_frame_rate(sample_rate).get_array_of_samples()
+    sound = wave.open(str(file), "rb")
+    nchannels = sound.getnchannels()
+    N = sound.getnframes()
+    dstr = sound.readframes(N*nchannels)
+    data = np.fromstring(dstr, np.int16)
+    data = data.reshape(-1, nchannels)
 
-    return np.frombuffer(sound, np.int16).flatten().astype(np.float32) / 32768.0
+    return data.flatten().astype(np.float32) / 32768.0
 
 def test_invalid_models():
     with p.raises(RuntimeError):
@@ -73,7 +77,7 @@ def test_from_params_file(models: str):
 def test_load_wav_file():
     np.testing.assert_almost_equal(
         preprocess(JFK_WAV),
-        w.api.load_wav_file(str(JFK_WAV.resolve())).mono,
+        w.api.load_wav_file_mono(str(JFK_WAV.resolve())),
     )
 
 def transcribe_strict():