ESPnet
The end-to-end speech processing toolkit behind a decade of ASR research baselines.
About
ESPnet is an end-to-end speech processing toolkit started in December 2017, built around Kaldi-style all-in-one recipes with a PyTorch backend (originally Chainer as well). It covers ASR, TTS, speech translation, enhancement, separation and diarization with reproducible recipes, and at roughly 10,000 GitHub stars and 27,000+ commits under Apache 2.0 it is still actively developed. Its recipe-published results on LibriSpeech, AISHELL and other corpora made it the default baseline that end-to-end ASR papers compare against.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Recipe-published results across LibriSpeech, AISHELL and other corpora; widely used as a research baseline (community-run).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 9/10
Scored 9/10 (verified tier, metrics-v1). Evidence: espnet/espnet repo verified active with ~10k stars and Apache-2.0 via GitHub API; recipe corpus results documented in the repo. Strengths: the field's standard E2E baseline, broad task coverage, sustained maintenance since 2017. Docked for: (1) research-oriented — production deployment takes assembly work. 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
- 8.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · ~9,985 stars; created Dec 2017; 27k+ commits
- · Kaldi-style recipes; PyTorch backend; ASR/TTS/ST/enhancement/diarization
- · Default E2E-ASR research baseline with recipe-published corpus results