I would like to know how do Kaldi and DeepSpeech speech recognition systems differ algorithmically? Which one would be more accurate for continuous speech in time?


Both of them using the end-to-end approach for speech recognition. However, because of the code complexity in DeepSpeech, you can't tune the model for your work. Kaldi could be configured in a different manner and you have access to the details of the models and indeed it is a modular tool. I think Kaldi could be a better tool academically and also commercially. But, Deepspeech is a BlackBox and could be a proper tool if your work is near to the work of DeepSpeech.

Moreover, if you are working in a language that there is not enough data for learning, you have a dozen of tools such as grapheme-to-phoneme to establish your dataset to start the learning process. Also, you can start with other models of Kaldi to work on which do not depend on the big data to be learned. However, DeepSpeech requires many hours of samples that is not a common asset for everyone in every language.

Also, you can know more about the traditional technique of speech processing vs the contemporary technique (deep learning) in this medium post.

  • $\begingroup$ thanks a lot for your reply. As you touched the subject deeply, is it possible to say the difference between them algorithmically like GMM/HMM or LSTM/RNN models they use. how to do they differ here? Thanks once again for such depth. $\endgroup$
    – Hanu
    Nov 26 '19 at 4:14
  • $\begingroup$ @Ramki my pleasure. I think this link: medium.com/@jonathan_hui/… could be useful. $\endgroup$
    – OmG
    Dec 1 '19 at 14:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.