Use FFmpeg6 in unittest (#3570)

Summary: Pull Request resolved: https://github.com/pytorch/audio/pull/3570

Reviewed By: huangruizhe

Differential Revision: D48518568

Pulled By: mthrok

fbshipit-source-id: 0fdfb8b3988789c7ded0fb336824034bedf6a394

.github/scripts/unittest-linux/install.sh

@@ -75,7 +75,7 @@ fi
 # Note: installing librosa via pip fail because it will try to compile numba.
 (
     set -x
-    conda install -y -c conda-forge ${NUMBA_DEV_CHANNEL} sox libvorbis 'librosa==0.10.0' parameterized 'requests>=2.20' 'ffmpeg>=5,<7'
+    conda install -y -c conda-forge ${NUMBA_DEV_CHANNEL} sox libvorbis 'librosa==0.10.0' parameterized 'requests>=2.20' 'ffmpeg>=6,<7'
     pip install kaldi-io SoundFile coverage pytest pytest-cov 'scipy==1.7.3' expecttest unidecode inflect Pillow sentencepiece pytorch-lightning 'protobuf<4.21.0' demucs tinytag pyroomacoustics flashlight-text git+https://github.com/kpu/kenlm
 )
 # Install fairseq