Issues

• Can audio codecs designed for reconstruction (e.g., DAC) be used as upstream models in s3prl?

  s3prl/s3prl#585 · Filter by author liusanchun5-art opened on Mar 15, 2026

• Upstream Trainable hugely reduce accuracy score for SID for both hubert and w2v2

  s3prl/s3prl#583 · Filter by author david-wei-01001 opened on Feb 14, 2026

• Masking issue in AttentivePoolingModule

  s3prl/s3prl#579 · Filter by author kpeverson opened on Jan 29, 2026

• Downstream checkpoints that lack reproducibility.

  s3prl/s3prl#577 · Filter by author hhhhhh45 opened on Jan 11, 2026

• Support for Continued Pretraining of wav2vec2 on New Unlabeled Audio

  s3prl/s3prl#576 · Filter by author akbar20gh opened on Dec 19, 2025

• Issues with deprecated api in torchaudio

  s3prl/s3prl#575 · Filter by author ftshijt opened on Dec 9, 2025
  3

• Leaderboard website is down

  s3prl/s3prl#574 · Filter by author mxmpl opened on Aug 20, 2025

• Audio Deepfake Detection Task

  s3prl/s3prl#573 · Filter by author hashim19 opened on Jul 29, 2025

• Inference for emotion is not implemented

  s3prl/s3prl#572 · Filter by author KansaiTraining opened on Jul 4, 2025

• Supports for Numpy 2.0 and above

  s3prl/s3prl#571 · Filter by author ftshijt opened on May 29, 2025
  3

• extra tokens are not optimised by CTC loss but used as targets for model to predict

  s3prl/s3prl#570 · Filter by author wenjie-p opened on May 2, 2025

• Use custom hubert as teacher to train a hubert distilled

  s3prl/s3prl#569 · Filter by author eloi-tavus opened on Apr 26, 2025
  1

• No Access to Fluent Speech Commands Dataset

  s3prl/s3prl#567 · Filter by author omertalmi5 opened on Mar 9, 2025

• Finetune the upstream

  s3prl/s3prl#561 · Filter by author LIN12955 opened on Dec 9, 2024

• Finetuning Distil Hubert Model for emotion recognition

  s3prl/s3prl#558 · Filter by author staru09 opened on Sep 12, 2024

• Looking for a Model Zoo for Downstream Tasks

  s3prl/s3prl#557 · Filter by author twmht opened on Sep 10, 2024
  1

• Performance mismatch between the train and inference phases (Downstream task: Speech Enhancement)

  s3prl/s3prl#552 · Filter by author seorim0 opened on Aug 23, 2024

• A way to save and load the output of upstream model for speedup

  s3prl/s3prl#546 · Filter by author shreyas2206 opened on Jun 19, 2024
  3

• run vq_apc pretrain failed

  s3prl/s3prl#542 · Filter by author moodeerf opened on Jun 4, 2024
  3

  

• How to extract weighted sum SSL representations from an audio dataset?

  s3prl/s3prl#539 · Filter by author mukherjeesougata-eros opened on May 6, 2024

• What is the accept rate in the VC task evaluation output?

  s3prl/s3prl#530 · Filter by author raotnameh opened on Feb 5, 2024
  1

• ValueError: mutable default <class 's3prl.upstream.roberta.roberta_model.EncDecBaseConfig'> for field encoder is not allowed: use default_factory

  s3prl/s3prl#522 · Filter by author haoyunlf opened on Dec 21, 2023
  6

• Different upstream and downstream learning rates

  s3prl/s3prl#521 · Filter by author xt0r3-Cambridge opened on Dec 14, 2023
  1

• University Work: Mispronunciation Detection

  s3prl/s3prl#508 · Filter by author xt0r3-Cambridge opened on Oct 5, 2023
  3

• support whisper as upstream models

  s3prl/s3prl#494 · Filter by author talallc opened on Jul 14, 2023
  1