fix(release): install MP3 tools before audio generation

Signed-off-by: Deluan <deluan@navidrome.org>

.github/workflows/release-audio-tests.yml

@@ -22,4 +22,8 @@ jobs:
       - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
         with:
           python-version: '3.12'
+      - name: Install tools for the offline MP3 decode test
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends ffmpeg
       - run: python3 -m unittest discover -s release/audio -p 'test_*.py' -v

.github/workflows/release-audio.yml

@@ -58,6 +58,11 @@ jobs:
       - uses: actions/setup-python@ece7cb06caefa5fff74198d8649806c4678c61a1 # v6
         with:
           python-version: '3.12'
+      - name: Install MP3 inspection tools before generation
+        if: env.AUDIO_MODE == 'audio' && env.AUDIO_ENABLED == 'true'
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y --no-install-recommends ffmpeg
       - name: Resolve sources and check limits and duplicate attempts
         id: prepare
         env:

release/audio/README.md

@@ -164,8 +164,9 @@ The suite blocks unmocked network access. Fixtures preserve the public bodies
 of v0.64.0/.1/.2 and test consequential omissions, unsafe tags/output, disabled
 generation, exact event identity, changing notes, duplicates/expiry/force,
 budget caps, error responses, and no retries. The separate PR test workflow
-never receives `OPENAI_API_KEY`. Generation uses only Python's standard library;
-the Ubuntu runner must have `ffprobe` and `ffmpeg` before speech is requested.
+never receives `OPENAI_API_KEY`. Generation uses only Python's standard library.
+Both workflows install `ffprobe` and `ffmpeg` in a credential-free step; audio
+generation installs them before source preparation or any OpenAI request.
 
 ## References