whisperlivekit/whisper_streaming_custom/backends.py: separator should be an empty space

whisperlivekit/whisper_streaming_custom/backends.py: use BatchedInferencePipeline in FasterWhisperASR class, and batch_size=16 in transcribe() function, also created WhisperXASR class

Allow CTranslate2 backend to choose device and compute types.

Corrects 'TranscriptionSegment' object is not subscriptable

solve #100

script to lib