Transcribe, Translate, or Transliterate: An Investigation of Intermediate Representations in Spoken Language Models

• URL: http://arxiv.org/abs/2510.02569v2
• Date: Thu, 16 Oct 2025 20:31:59 GMT
• Title: Transcribe, Translate, or Transliterate: An Investigation of Intermediate Representations in Spoken Language Models
• Authors: Tolúlopé Ògúnrèmí, Christopher D. Manning, Dan Jurafsky, Karen Livescu,
• Abstract summary: Spoken language models (SLMs) that integrate speech with large language models (LMs) rely on modality adapters (MAs) to map the output of speech encoders to a representation that is understandable to the decoder LM.<n>Here we examine the MA output representation in three SLMs (SALMONN, Qwen2-Audio and Phi-4-Multimodal-Instruct)<n>By finding the nearest decoder LM token to an MA representation, we uncover two strategies for MA representations.
• Score: 68.69744941948986
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Spoken language models (SLMs) that integrate speech with large language models (LMs) rely on modality adapters (MAs) to map the output of speech encoders to a representation that is understandable to the decoder LM. Yet we know very little about how these crucial MAs transform representations. Here we examine the MA output representation in three SLMs (SALMONN, Qwen2-Audio and Phi-4-Multimodal-Instruct). By finding the nearest decoder LM token to an MA representation, we uncover two strategies for MA representations. For models using a Whisper encoder, MAs appear to represent the meaning of the input using an English-based interlingua, allowing them to handle languages unseen in instruction tuning. For models that don't, like Phi-4-Multimodal-Instruct, MAs instead represent the phonetics of the input, but expressed with English words. We hypothesise that which arises depends on whether the speech encoder is trained only for speech recognition or also for translation.

Related papers

• What Makes a Good Speech Tokenizer for LLM-Centric Speech Generation? A Systematic Study [58.55905182336196]
  Speech-language models (SLMs) offer a promising path toward unifying speech and text understanding and generation.<n>We investigate the role of speech tokenizer designs in LLM-centric SLMs, augmented by speech heads and speaker modeling.<n>We introduce multi-token prediction (MTP) into SLMs, enabling each hidden state to decode multiple speech tokens.
  arXiv  Detail & Related papers  (2025-06-14T15:26:31Z)
• DC-Spin: A Speaker-invariant Speech Tokenizer for Spoken Language Models [45.791472119671916]
  Spoken language models (SLMs) process text and speech, enabling simultaneous speech understanding and generation. DC-Spin aims to improve speech tokenization by bridging audio signals and SLM tokens. We propose a chunk-wise approach to enable streamable DC-Spin without retraining and degradation.
  arXiv  Detail & Related papers  (2024-10-31T17:43:13Z)
• DM-Codec: Distilling Multimodal Representations for Speech Tokenization [18.779159914276573]
  Speech-language models have yielded significant improvements in speech tokenization and synthesis.<n>Recent efforts have unified acoustic and semantic tokens for improved performance, but overlook the crucial role of contextual representation.<n>We propose two novel distillation approaches: (1) a language model (LM)-guided distillation method that incorporates contextual information, and (2) a combined LM and self-supervised speech model (SM)-guided distillation technique.<n> Experiments show DM-Codec significantly outperforms state-of-the-art speech tokenization models, reducing WER by up to 13.46%, WIL by 9.82%, and improving speech quality by 5.
  arXiv  Detail & Related papers  (2024-10-19T07:14:14Z)
• Large Language Model Can Transcribe Speech in Multi-Talker Scenarios with Versatile Instructions [68.98811048970963]
  We present a pioneering effort to investigate the capability of large language models (LLMs) in transcribing speech in multi-talker environments.<n>We use WavLM and Whisper encoder to extract multi-faceted speech representations that are sensitive to speaker characteristics and semantic context.<n>Experiments reveal the promising performance of our proposed system, MT-LLM, in cocktail party scenarios.
  arXiv  Detail & Related papers  (2024-09-13T07:28:28Z)
• Cascaded Cross-Modal Transformer for Audio-Textual Classification [30.643750999989233]
  We propose to harness the inherent value of multimodal representations by transcribing speech using automatic speech recognition (ASR) models. We thus obtain an audio-textual (multimodal) representation for each data sample. We were declared the winning solution in the Requests Sub-Challenge of the ACM Multimedia 2023 Computational Paralinguistics Challenge.
  arXiv  Detail & Related papers  (2024-01-15T10:18:08Z)
• AudioPaLM: A Large Language Model That Can Speak and Listen [79.44757696533709]
  We introduce AudioPaLM, a large language model for speech understanding and generation. AudioPaLM fuses text-based and speech-based language models. It can process and generate text and speech with applications including speech recognition and speech-to-speech translation.
  arXiv  Detail & Related papers  (2023-06-22T14:37:54Z)
• Bidirectional Representations for Low Resource Spoken Language Understanding [39.208462511430554]
  We propose a representation model to encode speech in bidirectional rich encodings. The approach uses a masked language modelling objective to learn the representations. We show that the performance of the resulting encodings is better than comparable models on multiple datasets.
  arXiv  Detail & Related papers  (2022-11-24T17:05:16Z)
• LAMASSU: Streaming Language-Agnostic Multilingual Speech Recognition and Translation Using Neural Transducers [71.76680102779765]
  Automatic speech recognition (ASR) and speech translation (ST) can both use neural transducers as the model structure. We propose LAMASSU, a streaming language-agnostic multilingual speech recognition and translation model using neural transducers.
  arXiv  Detail & Related papers  (2022-11-05T04:03:55Z)
• Towards Language Modelling in the Speech Domain Using Sub-word Linguistic Units [56.52704348773307]
  We propose a novel LSTM-based generative speech LM based on linguistic units including syllables and phonemes. With a limited dataset, orders of magnitude smaller than that required by contemporary generative models, our model closely approximates babbling speech. We show the effect of training with auxiliary text LMs, multitask learning objectives, and auxiliary articulatory features.
  arXiv  Detail & Related papers  (2021-10-31T22:48:30Z)
• Fused Acoustic and Text Encoding for Multimodal Bilingual Pretraining and Speech Translation [21.622039537743607]
  We propose a Fused Acoustic and Text Masked Language Model (FAT-MLM) which jointly learns a unified representation for both acoustic and text in-put. Experiments on three translation directions show that our proposed speech translation models fine-tuned from FAT-MLM substantially improve translation quality.
  arXiv  Detail & Related papers  (2021-02-10T22:53:40Z)
• Bridging the Modality Gap for Speech-to-Text Translation [57.47099674461832]
  End-to-end speech translation aims to translate speech in one language into text in another language via an end-to-end way. Most existing methods employ an encoder-decoder structure with a single encoder to learn acoustic representation and semantic information simultaneously. We propose a Speech-to-Text Adaptation for Speech Translation model which aims to improve the end-to-end model performance by bridging the modality gap between speech and text.
  arXiv  Detail & Related papers  (2020-10-28T12:33:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.