PulseAugur
EN
LIVE 03:19:35

New REDDIT framework corrects ASR timestamp drift without model forgetting

By PulseAugur Editorial · [2 sources] ·

Researchers have developed REDDIT, a novel post-training framework designed to correct timestamp drift in Automatic Speech Recognition (ASR) systems without causing catastrophic forgetting. This method uses a replay-based distribution editing technique to refine timestamps by replaying the model's own decoder context while preserving non-timestamp token distributions. The framework successfully improved long-gap mIoU from 38.7% to 95.0% on Whisper-tiny with minimal parameter updates, while also significantly reducing out-of-domain timestamp errors. AI

IMPACT Improves accuracy and reliability of timestamping in ASR systems, potentially benefiting applications requiring precise temporal alignment.

RANK_REASON The cluster contains a research paper detailing a new method for improving ASR systems.

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →

New REDDIT framework corrects ASR timestamp drift without model forgetting

How we ranked this

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Research
The cluster contains a research paper detailing a new method for improving ASR systems.
Source corroboration
2 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
Topics
paper, model release
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
107 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.

Full methodology in our editorial standards.

COVERAGE [2]

  1. arXiv cs.AI TIER_1 English(EN) · Cheng-Kang Chou, Ming-To Chuang, Ke-Han Lu, Chan-Jan Hsu, Hung-yi Lee ·

    REDDIT: Correcting Model-Generated Timestamp Drift in ASR without Forgetting via Replay-Based Distribution Editing

    arXiv:2607.05364v1 Announce Type: cross Abstract: Modern autoregressive ASR systems can emit timestamps as decoded tokens, enabling timestamped transcription without frame-level aligners or inference-time post-processing. We show that these generated timestamps can drift across l…

  2. arXiv cs.AI TIER_1 English(EN) · Hung-yi Lee ·

    REDDIT: Correcting Model-Generated Timestamp Drift in ASR without Forgetting via Replay-Based Distribution Editing

    Modern autoregressive ASR systems can emit timestamps as decoded tokens, enabling timestamped transcription without frame-level aligners or inference-time post-processing. We show that these generated timestamps can drift across long non-speech spans: the transcript may remain pl…

RELATED ENTITIES

RELATED TOPICS