Pith. sign in

REVIEW 1 cited by

Investigating End-to-End ASR Architectures for Long Form Audio Transcription

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2309.09950 v2 pith:FT3TF7LP submitted 2023-09-18 eess.AS cs.SD

Investigating End-to-End ASR Architectures for Long Form Audio Transcription

classification eess.AS cs.SD
keywords modelsaudioconvolutionallongmodelarchitecturesattentioncategory
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

This paper presents an overview and evaluation of some of the end-to-end ASR models on long-form audios. We study three categories of Automatic Speech Recognition(ASR) models based on their core architecture: (1) convolutional, (2) convolutional with squeeze-and-excitation and (3) convolutional models with attention. We selected one ASR model from each category and evaluated Word Error Rate, maximum audio length and real-time factor for each model on a variety of long audio benchmarks: Earnings-21 and 22, CORAAL, and TED-LIUM3. The model from the category of self-attention with local attention and global token has the best accuracy comparing to other architectures. We also compared models with CTC and RNNT decoders and showed that CTC-based models are more robust and efficient than RNNT on long form audio.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. TRADE: Transducer-Augmented Decoder for Speech LLM

    cs.CL 2026-06 unverdicted novelty 6.0

    TRADE augments multimodal Speech LLMs with a transducer branch for streaming ASR, reporting 6.71% WER offline and 8.40% streaming on the Open ASR Leaderboard from one checkpoint.