The zamia speech project provides completely free (open source) tools and data for cloudless automatic speech recognition.

Everything here is free, open and transparent:

  • only open source speech recognition systems such as Kaldi-ASR, wav2letter++, DeepSpeech are used
  • all models are available for download for free
  • only freely available corpora are being used for model training
  • all metadata (reviewed transcripts, dictionaries, word replacement tables etc.) are free and open source as well

To find out more, please visit: GitHub project page

Model Overview


Model ASR Engine Language Hours Dict WER
20190609 tdnn_250 kaldi English 1500 160k 10.64%
20190609 tdnn_f kaldi English 1500 160k 8.84%
20190609 tdnn_fl kaldi English 1500 160k 5.80%
20190328 tdnn_250 kaldi German 412 450k 11.52%
20190328 tdnn_f kaldi German 412 450k 8.44%
20190427 w2l wav2letter German 412 450k 3.97%

French Models

French models by Paul Guyot can be found here:

Zamia Speech Logo