search

opea-project / GenAIComps

GenAI components at micro-service level; GenAI service composer to create mega-service
Apache License 2.0
78 stars 140 forks source link

[Feature] combine audio2text with asr service #905

Open lianhao opened 2 weeks ago

lianhao commented 2 weeks ago

Priority

Undecided

OS type

Ubuntu

Hardware type

Xeon-GNR

Running nodes

Single Node

Description

The audio2text service is almost the same as the asr service, except for the returned data object type. We should combine both of them to minimize maintenance effort

Spycsh commented 1 week ago

Hi @lianhao due to a legacy issue https://github.com/huggingface/tokenizers/issues/187, the first implementation of these ASR/TTS code is to split the microservice start and the model initialization to different places. Basically it is because our microservice are initialized in the other process using register_microservice and the huggingface tokenizer just fail with that case.

It is a bit annoying before I finally get some solutions to avoid such situations. I fully agree we should combine these two services together and remove the wrapper, same to tts service. I am starting to look into this together with the OpenAI format compatible feature.