AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Textually Pretrained Speech Language Models

Michael Hassid, Tal Remez, Tu Anh Nguyen, Itai Gat, Alexis Conneau, Felix Kreuk, Jade Copet, Alexandre Défossez, Gabriel Synnaeve, Emmanuel Dupoux, Roy Schwartz, Yossi Adi

NeurIPS 2023 · Large language models › Language, safety and interpretability · approach: Speech and audio language models · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Warm-initialize the speech model from pretrained OPT-350M rather than random initializationSpeechLM · Zero-shot speech modeling, 100/200/500 tokens at 50Hz and 500 tokens at 25HzsWUGGY: 68.91 -> 71.30; 69.85 -> 72.92; 66.65 -> 70.69; 79.44 -> 81.42Improved
Use a lower speech-token frequency, 25Hz instead of 50Hz (ablation)OPT-350M SpeechLM · Zero-shot speech modelingsWUGGY: 70.69 -> 81.42Improved
Increase the speech tokenizer vocabulary from 100 to 200 tokens at 50Hz (ablation)OPT-350M SpeechLM · Zero-shot speech modelingsWUGGY: 71.30 -> 72.92Improved
Increase the speech tokenizer vocabulary from 200 to 500 tokens at 50Hz (ablation)OPT-350M SpeechLM · Zero-shot speech modelingsWUGGY: 72.92 -> 70.69Hurt
Warm-initialize using pretrained textual OPT rather than random initializationSpeechLM · Speech generationtext-PPL: 88.17 -> 74.95; 117.17 -> 92.69; 67.68 -> 47.8Improved
Increase model size from 1.3B to 13B parameters (ablation)TWIST · sStoryClozeaccuracy: 52.4 -> 55.4Improved
Increase model size from 1.3B to 13B parameters (ablation)TWIST · tStoryClozeaccuracy: 70.6 -> 76.4Improved
Increase model size from 1.3B to 7B parameters (ablation)TWIST · LibriSpeech-cleanMMOS: 3.51($\pm$0.07) -> 3.79($\pm$0.06)Improved
Initialize speech model from image-pretrained ImageGPT instead of random initialization (ablation)SpeechLM · ImageGPT initialization, 200 tokens, 50HzPPL: 5.22 -> 8.21Hurt
Initialize speech model from image-pretrained ImageGPT instead of random initialization (ablation)SpeechLM · ImageGPT initialization, 200 tokens, 50HzsWUGGY: 72.22 -> 54.26Hurt
Initialize speech model from image-pretrained ImageGPT instead of random initialization (ablation)SpeechLM · ImageGPT initialization, 500 tokens, 25HzsBLIMP: 56.39 -> 51.84Hurt
Use pretrained textual BLOOM or Pythia weights for warm initializationSpeechLM · Zero-shot speech modelingsBLIMP: BLOOM: 53.03 -> 55.79; Pythia: 53.07 -> 56.00Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy