1 # FFmpeg TensorFlow integration 2 3 Decoding audio files can be done using a new op that uses 4 [FFmpeg](http://www.ffmpeg.org) to convert various audio file formats into 5 tensors. 6 7 tf.audio.decode_audio accepts MP3, WAV, and OGG file formats. 8 9 FFmpeg must be installed before these ops can be used. The ops will look for the 10 ffmpeg binary somewhere in `$PATH`. When the binary is unavailable, the error 11 `FFmpeg must be installed to run this op. FFmpeg can be found at 12 http://www.ffmpeg.org.` will be returned. 13 14 ## Testing 15 16 In addition to the regular tests, the integration tests should also be 17 run on this code. First, install `docker`. Then run the integration tests: 18 19 ```shell 20 export TF_BUILD_CONTAINER_TYPE=CPU # or GPU 21 export TF_BUILD_PYTHON_VERSION=PYTHON2 # or PYTHON3 22 export TF_BUILD_IS_OPT=OPT 23 export TF_BUILD_IS_PIP=PIP 24 export TF_BUILD_INTEGRATION_TESTS=1 25 tensorflow/tools/ci_build/ci_parameterized_build.sh 26 ``` 27