Skip to content
← Models
Open weightsSpeechBrain/10 trustVerified

SpeechBrain

The all-in-one PyTorch speech toolkit: 200+ recipes and pretrained models for ASR and everything around it.

Visit official page →

About

SpeechBrain is an open-source PyTorch toolkit launched in March 2021 that has become one of the most-used speech frameworks: roughly 11,900 GitHub stars, and the SpeechBrain 1.0 paper cites 2.5 million monthly downloads. It ships 200+ training recipes across 40+ datasets covering ASR, speaker verification, speech enhancement, separation, diarization and TTS, with pretrained models on Hugging Face and a consistent YAML-hyperparameter training interface. Its ASR recipes publish reproducible LibriSpeech results, which is why it doubles as a standard baseline in speech research. Released under Apache 2.0.

Specs

Context window

Model-dependent; recipe-configured

Pricing

Free (open source)

Free quota

Free and open source under Apache-2.0

API access

Self-hosted library (pip install speechbrain); pretrained models via Hugging Face

Weights

Open weights

Released

Not disclosed

Benchmarks

Community-run, reproducible recipe results (e.g. LibriSpeech ASR recipes); no single vendor leaderboard — figures come from the published recipes.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

asr-researchspeech-recognition-trainingspeaker-verificationspeech-enhancement

Trust score — 9/10

/10 trustVerified

Scored 9/10 (verified tier, metrics-v1). Evidence: speechbrain/speechbrain repo verified active (commits within days) with ~11.9k stars and Apache-2.0 via GitHub API; SpeechBrain 1.0 paper (arXiv 2407.00463) documents scale and design. Strengths: huge recipe coverage, reproducible results, healthy maintenance. Docked for: (1) framework-level entry — quality claims belong to individual recipes, not one model. License: Apache-2.0.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
9.0/10
  • · ~11,870 stars; launched March 2021; commits within days
  • · 200+ recipes across 40+ datasets per SpeechBrain 1.0 paper (arXiv 2407.00463)
  • · Pretrained models on Hugging Face; YAML-hyperparameter training interface
  • · Reproducible LibriSpeech ASR recipe results

Sources