Avihu Dekel

The architecture, training methodology and inference speedups of Granite 5.0 Turbo CTC result in a model that is on the speed-accuracy Pareto frontier of the Open ASR leaderboard for English short-form ASR while being twice as fast as the fastest competitor.

This paper presents models that add timestamps and speaker information to a given transcript using a non-autoregressive LLM-based architecture and achieves state-of-the-art timestamp accuracy and the best cpWER for speaker attribution.

Speaker-attributed ASR gets a serious boost from jointly training speaker cluster tags within a speech-aware LLM, outperforming traditional pipelines.

Despite advances in expressive speech, current TTS systems often miss subtle but crucial contextual cues, failing to emphasize the correct words even when the context makes the intended meaning clear.

Skip reinforcement learning and still get SOTA vision-language reasoning performance with a simple loss re-weighting scheme that cuts training time by 7x.

LLM-based ASR can be sped up by 4.4x with minimal accuracy loss by using a CTC encoder to speculatively generate draft transcriptions.

Ditch slow, sequential decoding: NLE achieves 27x speedup over autoregressive ASR by using a non-autoregressive, LLM-based transcript editing approach.