[KongchangAI]
ConceptToken-and-Duration Transducer

TDT

A speech recognition decoder architecture that jointly models tokens and durations, improving alignment and recognition performance.

Timeline (last 90 days)

Nov 11

When sharing an encoder and jointly training the CTC and TDT losses, the TDT-CTC hybrid architecture can combine the fast convergence of CTC with the fine alignment capability of TDT, and typically outperforms a single architecture in multilingual, variable-speech-rate scenarios.

Unverified50%
Nov 3

ufakzeka-kulak is an open-source Turkish speech-to-text model with only 31.6M parameters, using a FastConformer encoder and TDT and CTC heads, trained from scratch, and released under the Apache-2.0 license.

Unverified50%
Nov 3

With greedy TDT decoding, the model achieves a normalized WER of 0.0921 on the FLEURS Turkish test set, with a 95% confidence interval of 0.0792 to 0.1055.

Unverified50%

All Facts (3)

Source Articles