# s3prl/s3prl

Self-Supervised Speech Pre-training and Representation Learning Toolkit

Repository: https://github.com/s3prl/s3prl
Canonical: https://ross.abutalabs.com/products/s3prl
Homepage: https://s3prl.github.io/s3prl/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: speech-representation, mockingjay, representation-learning, apc, tera, self-supervised-learning, speech-pretraining, vq-apc, wav2vec, vq-wav2vec, wav2vec2, cpc, pase, decoar, hubert, distilhubert, wavlm, unispeech-sat, decoar2, data2vec
Last push: 2026-03-12T19:14:49+00:00

## Health v2 (maintenance only)
Score: 55/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 71, release rhythm 8, longevity 100
- inputs: {"age_days": 2607, "days_push": 174, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2561, forks 535 (observed 2026-08-28T04:07:00.965113+00:00)

## What it is
S3PRL is a PyTorch toolkit for self-supervised speech pre-training and representation learning, bundling many upstream models like wav2vec 2.0, HuBERT, WavLM, and Mockingjay. It is now in pure maintenance mode, keeping existing functions working while accepting only new upstream model contributions.

## Use cases
- extract speech representations from pretrained self-supervised models
- fine-tune wav2vec 2.0 or HuBERT for speech recognition
- benchmark speech models on SUPERB tasks
- pre-train speech models like APC or TERA
- evaluate embeddings for speaker verification and emotion recognition
- run speech downstream tasks like ASR, keyword spotting, and diarization

## When to choose
- you need a unified interface to many self-supervised speech models
- you want to reproduce SUPERB benchmark results
- you need pretrained speech encoders for downstream speech tasks

## When to avoid
- you need new features beyond upstream models, since the project is in maintenance mode
- you want a lightweight production ASR pipeline rather than a research toolkit
- you need non-speech or text-only self-supervised learning

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, speech-recognition, audio-processing, deep-learning
- domain: speech-processing, machine-learning, deep-learning
- platform: python
- tags: self-supervised-learning, speech-pretraining, wav2vec, hubert, representation-learning, pytorch, superb, audio, linux, gpu

## Member repositories
- s3prl/s3prl (main) score 55

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:00.965113+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:23:30.577403+00:00, confidence not recorded.
  - readme: https://github.com/s3prl/s3prl (fetched 2026-08-28T04:07:00.965113+00:00, sha 4f920d4f0014)
  - homepage: https://s3prl.github.io/s3prl/ (fetched 2026-08-29T10:06:05.827261+00:00, sha efb575757ade)
  - registry_pypi: https://pypi.org/pypi/s3prl/json (fetched 2026-08-29T10:06:05.837152+00:00, sha 5b668dbe57b9)
- Data as of 2026-08-30T08:39:29.467469+00:00.