eduzhai > Applied Sciences > Engineering >

Quantization of Acoustic Model Parameters in Automatic Speech Recognition Framework

  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: State-of-the-art hybrid automatic speech recognition (ASR) system exploitsdeep neural network (DNN) based acoustic models (AM) trained with LatticeFree-Maximum Mutual Information (LF-MMI) criterion and n-gram language models.The AMs typically have millions of parameters and require significant parameterreduction to operate on embedded devices. The impact of parameter quantizationon the overall word recognition performance is studied in this paper. Followingapproaches are presented: (i) AM trained in Kaldi framework with conventionalfactorized TDNN (TDNN-F) architecture, (ii) the TDNN AM built in Kaldi loadedinto the PyTorch toolkit using a C++ wrapper for post-training quantization,(iii) quantization-aware training in PyTorch for Kaldi TDNN model, (iv)quantization-aware training in Kaldi. Results obtained on standard Librispeechsetup provide an interesting overview of recognition accuracy w.r.t. appliedquantization scheme.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...