Spaces:

kotoba-speech

/

kotoba-whisper-demo

like 13

Running on Zero

Currently, the mic & file input takes the audio as is, but we need to convert the audio into 16kHz.

The pipeline does resampling when it receives a file path instead of array, and gradio audio pass the filepath to the pipeline so it is resampling to 16kHz already.
https://github.com/huggingface/transformers/blob/09f9f566de83eef1f13ee83b5a1bbeebde5c80c1/src/transformers/pipelines/automatic_speech_recognition.py#L361

https://huggingface.co/docs/transformers/v4.39.3/en/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline.__call__.inputs