add a new faster model to infer in cli

* update yaml and readme, test=tts

change the infer in order to implement the new faster model for text

Add a new model for faster text process in cli

Add a new model for faster text process

* Replaced the fixed path with the dynamic path in MFA, test=doc

* add finetune demos, test=doc

* update apt docs, test=doc

[s2t] update wenetspeech RESULT.md, add chunk=-1 result of online conformer

* finetune add frozen

Removed useless spk_id in speech_server and streaming_tts_server from demos, and support bilingual server engine, test=tts (#2380)

* Removed useless spk_id in speech_server and streaming_tts_server from demos, and support bilingual server engine.

* fix ernie_sat, test=tts

* fix for comments, test=tts

Co-authored-by: yuehuayingxueluo <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>

[s2t] DataParallel init method changed, fixed conformer could not multi-gpu training and don't affect dy2st

[s2t] update wenetspeech streaming coformer result

* Revised must_neural_tone_words.

add tool to compare two tests results, show difference