MétaCan
← all works

Internalizing ASR with Implicit Chain of Thought for Efficient Speech-to-Speech Conversational LLM

2024· preprint· en· 1 citations· W4403794801 on OpenAlex· 10.48550/arxiv.2409.17353

Why is this work in the frame?

A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.

Canadian funderA Canadian agency funded it. The work may carry no Canadian affiliation at all.

No Canadian affiliation. An affiliation-only frame — the usual design — would never have seen this work. It is one of the works that make the case for inverting the frame.

The three-model screen

all 1,000 screened works →

All three models called this out of scope.

stratum: fund_new · design weight: 1678.90 (the sample is stratified; any rate computed without the weight is wrong)
Claude Opus 4.8OUT
genre: empirical
about Canada: no
confidence: high

Machine learning method internalizing ASR chain of thought in a speech-to-speech conversational LLM; a model architecture contribution.

GPT-5.6 (high)OUT
genre: empirical
about Canada: no
confidence: high

The preprint develops a speech-to-speech language-model method, not a study of research practice.

Grok 4.5OUT
genre: empirical
about Canada: no
confidence: high

Speech LLM engineering for ASR chain-of-thought; AI systems research, not metaresearch.

Abstract

Current speech-based LLMs are predominantly trained on extensive ASR and TTS datasets, excelling in tasks related to these domains. However, their ability to handle direct speech-to-speech conversations remains notably constrained. These models often rely on an ASR-to-TTS chain-of-thought pipeline, converting speech into text for processing before generating audio responses, which introduces latency and loses audio features. We propose a method that implicitly internalizes ASR chain of thought into a speech LLM, enhancing its native speech understanding capabilities. Our approach reduces latency and improves the model's native understanding of speech, paving the way for more efficient and natural real-time audio interactions. We also release a large-scale synthetic conversational dataset to facilitate further research.

Stored with the screening record, where it is evidence for the labels above.

The record

Venue
arXiv (Cornell University)
Topic
Speech and dialogue systems
Field
Computer Science
Canadian institutions
Funders
Alliance de recherche numérique du CanadaNatural Sciences and Engineering Research Council of CanadaUniversity of British Columbia
Keywords
Speech recognitionPsychologyIndirect speechComputer scienceLinguisticsCognitive psychologyNatural language processingPhilosophy
Has abstract in OpenAlex
yes