Update README.md - ffmpeg dependency (#224)

* Update README.md - ffmpeg dependency

Computing metrics uses whisper for transcribing audio and Whisper has a dependency on ffmpeg.

* Fix typo

---------

Co-authored-by: Kaushik Ram Sadagopan <krs@fb.com>

README.md

@@ -62,6 +62,10 @@ The Seamless model is the unified model for expressive streaming speech-to-speec
 pip install .
 ```
 
+> [!NOTE]
+> Transcribing inference audio for computing metric uses [Whisper](https://github.com/openai/whisper#setup), which is automatically installed. Whisper in turn requires the command-line tool [`ffmpeg`](https://ffmpeg.org/) to be installed on your system, which is available from most package managers.
+
+
 ## Running inference
 
 ### SeamlessM4T Inference