Kaldi
The C++ speech recognition toolkit that defined a decade of ASR research.
About
Kaldi is the free, open-source C++ toolkit for speech recognition research, built around finite-state transducers (via OpenFst), extensive linear-algebra support and complete recipes that build working recognizers from public corpora. Started by Daniel Povey and collaborators around 2009-2011 and released under Apache 2.0, it gathered about 15,500 GitHub stars and became the standard platform new ASR ideas were compared against for a decade. Its recipes, nnet3 chain models and online decoding still show up in production systems, and most modern toolkits (ESPnet included) inherited its recipe style.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Recipe-published results across dozens of corpora; the long-standing community baseline rather than a single leaderboard number.
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: kaldi-asr/kaldi repo verified with ~15.5k stars; Apache-2.0 widely documented (GitHub API returns NOASSERTION on the license field, but the license is stated in-repo and in the toolkit paper); kaldi-asr.org live. Strengths: the field's foundational toolkit, enormous recipe corpus, proven in production. Docked for: (1) C++ legacy codebase with a steep learning curve; (2) slower-moving than the PyTorch-native toolkits. License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 10.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · ~15,498 stars; the ASR research standard for a decade
- · FST-based C++ toolkit (OpenFst) with complete public-corpus recipes
- · nnet3 chain models and online decoding still used in production
- · Active but slower-moving; C++ legacy codebase