Open Models Almanac
Methodsmodel architecturemultimodal architecture

specific method · filed under model architecture

Discrete audio token encoding

Inkling's "discrete token encoding" is dMel: each of 80 mel bins is quantized to one of 16 levels and the 80 embeddings are summed into one token every 50 ms, with no audio encoder or learned codebook.

Written for the explainer Hearing, speaking and keeping time: audio, omni models and multimodal positions.

Also called Discrete token encoding for audio, discrete token encoding, discrete token encoding for audio inputs.

sources
4
model
1
lab adopt it
1
strongest
core

Documented in

Thinking Machines Lab

Evidence

4 quotes from 4 sources, one card per model, each opening with its strongest.

How sources treat it
core 4

Inkling

Thinking Machines Labadopted4 sources

audio via discrete token encoding

coremodel architecturein InklingThinking Machines Lab
3 more from this model

audio via discrete token encoding

coremodel architecturein InklingThinking Machines Lab

audio via discrete token encoding

coremodel architecturein InklingThinking Machines Lab

audio via discrete token encoding

coremodel architecturein InklingThinking Machines Lab
↑