[!206][FIX] Fix speech requirements

# Why is the change needed?
Some code for speech processing in `fbk-fairseq` could not run without some packages which were not in the requirements.

# What changes does the patch introduce?
Added `soundfile`, `pandas`, `sentencepiece` to `speech_requirements.txt`.

# How was this patch tested?
Started a new virtual environment and checked that the updated requirements were sufficient to run e.g. `$FAIRSEQ_DIR/examples/speech_to_text/preprocess_generic.py`.

Author: Lina Conti

speech_requirements.txt

@@ -3,3 +3,6 @@ ctc_segmentation
 srt
 praat-parselmouth
 pangolinn
+soundfile
+pandas
+sentencepiece