search

TensorSpeech / TensorFlowASR

:zap: TensorFlowASR: Almost State-of-the-art Automatic Speech Recognition in Tensorflow 2. Supported languages that can use characters or subwords
https://huylenguyen.com/asr
Apache License 2.0
917 stars 242 forks source link

Contextual Biasing | Adding hints at runtime for Streaming transducers #108

Open tumusudheer opened 3 years ago

tumusudheer commented 3 years ago

Hi,

It is really great work. Thank you very much for the streaming transducers. Is it possible to add hints at runtime (In streaming transducers)(Section: 4) (Say I have some names which are not available at the time of training but at the inference time, I've some knowledge on what I'm looking for), How can I use a set/list of phrases as hints at inference (testing) time ?

Thank you

nglehuy commented 3 years ago

The contextual biasing is not supported yet :smile: May be after I figure out how to add LM to decoder

tumusudheer commented 3 years ago

Thanks @usimarit

The streaming demonstration example uses conformer. Do you have a sample code for streaming audio using streaming transducers ?

Also, I want to train streaming transducers. Seems only conformer pre-trained models are available #61. It would be great if you can provide streaming transducers pre-trained model to finetune from ?

Thank you

nglehuy commented 3 years ago

@tumusudheer No currently I haven't trained streaming transducer so I don't have sample code for streaming audio and pretrained model.