Work in progress for batching with multiple audio files

This work continues where SYSTRAN#1302 left off. The goal is to
transcribe multiple audio files truly in parallel and increase
GPU throughput.

Author: j-silv