Add resampling flag for wav conversion

.github/workflows/release.yml

@@ -10,12 +10,12 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout code
-        uses: actions/checkout@v2
+        uses: actions/checkout@v4
         with:
           lfs: true
       - name: Retrieve version
         run: |
-          echo "::set-output name=TAG_NAME::$(grep -i -o -P '(?<=version=\")[^\"]+(?=\")' setup.py)"
+          echo "name=TAG_NAME::$(grep -i -o -P '(?<=version=\")[^\"]+(?=\")' setup.py)" > "$GITHUB_OUTPUT"
         id: version
       - name: Changelog
         uses: scottbrenner/generate-changelog-action@master
@@ -24,10 +24,12 @@ jobs:
         id: create_release
         uses: softprops/action-gh-release@v1
         env:
-          GITHUB_TOKEN: ${{ secrets.GH_TOKEN }} # This token is provided by Actions, you do not need to create your own token
+          # This token is provided by Actions, you do not need to create your
+          # own token
+          GITHUB_TOKEN: ${{ secrets.GH_TOKEN }}
         with:
           tag_name: ${{ steps.version.outputs.TAG_NAME }}
-          release_name: Release ${{ steps.version.outputs.TAG_NAME }}
+          name: Release ${{ steps.version.outputs.TAG_NAME }}
           body: ${{steps.changelog.outputs.changelog}}
           draft: false
           prerelease: false

audiotools/core/ffmpeg.py

@@ -176,7 +176,14 @@ def load_from_file_with_ffmpeg(cls, audio_path: str, quiet: bool = True, **kwarg
 
             ff = ffmpy.FFmpeg(
                 inputs={audio_path: None},
-                outputs={wav_file: None},
+                # For inputs that are m4a (and others?), the input audio can
+                # have samples that don't match the sample rate. This aresample
+                # option forces ffmpeg to read timing information in the source
+                # file instead of assuming constant sample rate.
+                #
+                # This fixes an issue where an input m4a file might be a
+                # different length than the output wav file
+                outputs={wav_file: "-af aresample=async=1000"},
                 global_options=global_options,
             )
             ff.run()

setup.py

@@ -6,7 +6,7 @@
 
 setup(
     name="descript-audiotools",
-    version="0.7.3",
+    version="0.7.4",
     classifiers=[
         "Intended Audience :: Developers",
         "Intended Audience :: Education",

tests/core/test_whisper.py

@@ -26,8 +26,7 @@ def test_whisper_transcript():
 
     transcript = signal.get_whisper_transcript()
 
-    assert "<|startoftranscript|>" in transcript
-    assert "<|endoftext|>" in transcript
+    assert transcript.startswith("<|startoftranscript|>")
 
 
 def test_whisper_embeddings():