Audio Codes "at utterance level"

[juanilarregui]

❓ Questions

I'm interested in using the encoder to encode an audio fragment of a few seconds into just one codebook vector. However, the model returns a sequence of several audio_codes (of course, it is the only way to succesfully decode the audio afterwards).

How would you recommend using the encoder, and/or pre-postprocessing the audio input or audio_codes to obtain just one audio code "at utterance level"?

Thanks in advance.