Speech-FT: Merging Pre-trained And Fine-Tuned Speech Representation Models For Cross-Task Generalization

• URL: http://arxiv.org/abs/2502.12672v2
• Date: Sun, 25 May 2025 14:07:23 GMT
• Title: Speech-FT: Merging Pre-trained And Fine-Tuned Speech Representation Models For Cross-Task Generalization
• Authors: Tzu-Quan Lin, Wei-Ping Huang, Hao Tang, Hung-yi Lee,
• Abstract summary: Fine-tuning speech representation models can enhance performance on specific tasks but often compromises cross-task generalization ability.<n>Existing approaches, such as regularizing weight changes during fine-tuning, may fail to maintain sufficiently high feature similarity with the pre-trained model.<n>We propose Speech-FT, a novel two-stage fine-tuning framework designed to maintain cross-task generalization while benefiting from fine-tuning.
• Score: 51.56024241398741
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Fine-tuning speech representation models can enhance performance on specific tasks but often compromises their cross-task generalization ability. This degradation is often caused by excessive changes in the representations, making it difficult to retain information learned during pre-training. Existing approaches, such as regularizing weight changes during fine-tuning, may fail to maintain sufficiently high feature similarity with the pre-trained model, and thus could possibly lose cross-task generalization. To address this issue, we propose Speech-FT, a novel two-stage fine-tuning framework designed to maintain cross-task generalization while benefiting from fine-tuning. Speech-FT first applies fine-tuning specifically designed to reduce representational drift, followed by weight-space interpolation with the pre-trained model to restore cross-task generalization. Extensive experiments on HuBERT, wav2vec 2.0, DeCoAR 2.0, and WavLM Base+ demonstrate that Speech-FT consistently improves performance across a wide range of supervised, unsupervised, and multitask fine-tuning scenarios. Moreover, Speech-FT achieves superior cross-task generalization compared to fine-tuning baselines that explicitly constrain weight changes, such as weight-space regularization and LoRA fine-tuning. Our analysis reveals that Speech-FT maintains higher feature similarity to the pre-trained model compared to alternative strategies, despite allowing larger weight-space updates. Notably, Speech-FT achieves significant improvements on the SUPERB benchmark. For example, when fine-tuning HuBERT on automatic speech recognition, Speech-FT is able to reduce phone error rate from 5.17% to 3.94%, lower word error rate from 6.38% to 5.75%, and increase speaker identification accuracy from 81.86% to 84.11%. Speech-FT provides a simple yet powerful solution for further refining speech representation models after pre-training.

Related papers

• Selective Self-to-Supervised Fine-Tuning for Generalization in Large Language Models [24.659722730219134]
  This paper introduces Selective Self-to-Supervised Fine-Tuning (S3FT)<n>S3FT achieves better performance than the standard supervised fine-tuning (SFT) while improving generalization.<n>The effectiveness of S3FT is demonstrated through experiments on mathematical reasoning, Python programming and reading comprehension tasks.
  arXiv  Detail & Related papers  (2025-02-12T05:24:21Z)
• EMOVA: Empowering Language Models to See, Hear and Speak with Vivid Emotions [152.41217651729738]
  GPT-4o is an omni-modal model that enables vocal conversations with diverse emotions and tones. We propose EMOVA to enable Large Language Models with end-to-end speech capabilities. For the first time, EMOVA achieves state-of-the-art performance on both the vision-language and speech benchmarks.
  arXiv  Detail & Related papers  (2024-09-26T16:44:02Z)
• SimpleSpeech 2: Towards Simple and Efficient Text-to-Speech with Flow-based Scalar Latent Transformer Diffusion Models [64.40250409933752]
  We build upon our previous publication by implementing a simple and efficient non-autoregressive (NAR) TTS framework, termed SimpleSpeech 2. SimpleSpeech 2 effectively combines the strengths of both autoregressive (AR) and non-autoregressive (NAR) methods. We show a significant improvement in generation performance and generation speed compared to our previous work and other state-of-the-art (SOTA) large-scale TTS models.
  arXiv  Detail & Related papers  (2024-08-25T17:07:39Z)
• Controlling Whisper: Universal Acoustic Adversarial Attacks to Control Speech Foundation Models [3.1511847280063696]
  Speech enabled foundation models can perform tasks other than automatic speech recognition using an appropriate prompt. With the development of audio-prompted large language models there is the potential for even greater control options. We demonstrate that with this greater flexibility the systems can be susceptible to model-control adversarial attacks.
  arXiv  Detail & Related papers  (2024-07-05T13:04:31Z)
• DeSTA: Enhancing Speech Language Models through Descriptive Speech-Text Alignment [82.86363991170546]
  We propose a Descriptive Speech-Text Alignment approach that leverages speech captioning to bridge the gap between speech and text modalities. Our model demonstrates superior performance on the Dynamic-SUPERB benchmark, particularly in generalizing to unseen tasks. These findings highlight the potential to reshape instruction-following SLMs by incorporating descriptive rich, speech captions.
  arXiv  Detail & Related papers  (2024-06-27T03:52:35Z)
• SpeechGPT-Gen: Scaling Chain-of-Information Speech Generation [56.913182262166316]
  Chain-of-Information Generation (CoIG) is a method for decoupling semantic and perceptual information in large-scale speech generation. SpeechGPT-Gen is efficient in semantic and perceptual information modeling. It markedly excels in zero-shot text-to-speech, zero-shot voice conversion, and speech-to-speech dialogue.
  arXiv  Detail & Related papers  (2024-01-24T15:25:01Z)
• uSee: Unified Speech Enhancement and Editing with Conditional Diffusion Models [57.71199494492223]
  We propose a Unified Speech Enhancement and Editing (uSee) model with conditional diffusion models to handle various tasks at the same time in a generative manner. Our experiments show that our proposed uSee model can achieve superior performance in both speech denoising and dereverberation compared to other related generative speech enhancement models.
  arXiv  Detail & Related papers  (2023-10-02T04:36:39Z)
• SpeechGPT: Empowering Large Language Models with Intrinsic Cross-Modal Conversational Abilities [39.07096632751864]
  SpeechGPT is a large language model with intrinsic cross-modal conversational abilities. We employ a three-stage training strategy that includes modality-adaptation pre-training, cross-modal instruction fine-tuning, and chain-of-modality instruction fine-tuning.
  arXiv  Detail & Related papers  (2023-05-18T14:23:25Z)
• Robust Speech Recognition via Large-Scale Weak Supervision [69.63329359286419]
  We study the capabilities of speech processing systems trained simply to predict large amounts of transcripts of audio on the internet. When scaled to 680,000 hours of multilingual and multitask supervision, the resulting models generalize well to standard benchmarks. We are releasing models and inference code to serve as a foundation for further work on robust speech processing.
  arXiv  Detail & Related papers  (2022-12-06T18:46:04Z)
• Self-supervised Rewiring of Pre-trained Speech Encoders: Towards Faster Fine-tuning with Less Labels in Speech Processing [66.92823764664206]
  We take a sober look into pre-trained speech encoders and rewire their representation space without requiring task-specific labels. Our experiments on 6 speech processing tasks, exhibit a significant convergence speedup during task fine-tuning as well as consistent task improvement.
  arXiv  Detail & Related papers  (2022-10-24T08:27:09Z)
• TranSpeech: Speech-to-Speech Translation With Bilateral Perturbation [61.564874831498145]
  TranSpeech is a speech-to-speech translation model with bilateral perturbation. We establish a non-autoregressive S2ST technique, which repeatedly masks and predicts unit choices. TranSpeech shows a significant improvement in inference latency, enabling speedup up to 21.4x than autoregressive technique.
  arXiv  Detail & Related papers  (2022-05-25T06:34:14Z)
• WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing [102.45426364965887]
  We propose a new pre-trained model, WavLM, to solve full-stack downstream speech tasks. WavLM is built based on the HuBERT framework, with an emphasis on both spoken content modeling and speaker identity preservation. We scale up the training dataset from 60k hours to 94k hours of public audio data, and optimize its training procedure for better representation extraction.
  arXiv  Detail & Related papers  (2021-10-26T17:55:19Z)
• Performance-Efficiency Trade-offs in Unsupervised Pre-training for Speech Recognition [32.61769580342906]
  We focus on wav2vec 2.0, and formalize several architecture designs that influence both the model performance and its efficiency. We introduce SEW (Squeezed and Efficient Wav2vec), a pre-trained model architecture with significant improvements along both performance and efficiency dimensions.
  arXiv  Detail & Related papers  (2021-09-14T17:58:09Z)
• Towards Multi-Scale Style Control for Expressive Speech Synthesis [60.08928435252417]
  The proposed method employs a multi-scale reference encoder to extract both the global-scale utterance-level and the local-scale quasi-phoneme-level style features of the target speech. During training time, the multi-scale style model could be jointly trained with the speech synthesis model in an end-to-end fashion.
  arXiv  Detail & Related papers  (2021-04-08T05:50:09Z)

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.