Skip to content

Latest commit

 

History

History
54 lines (32 loc) · 1.26 KB

ESPnet-inference.md

File metadata and controls

54 lines (32 loc) · 1.26 KB
Raw

Install ESPnet for inference only

Install ESPnet and (optionally) Kaldi

    conda activate a1003
    pip install espnet torchaudio
    conda install kaldi

Download espnet code for review

    cd ~
    git clone https://github.com/espnet/espnet

Open ~/espent directory in VScode

Download ASR models

git lfs is required

    (a1003)$ conda install git-lfs

Download models from huggingface

    (a1003)$ cd a1003
    (a1003)$ mkdir -p models
    (a1003)$ cd models
    (a1003)$ git clone https://hf.co/pkyoung/ma16k2401a
    (a1003)$ git clone https://hf.co/pkyoung/ma16k2401b
    (a1003)$ git clone https://hf.co/pkyoung/ma16k2401c

Run inference

Check models, data and configuration

    (a1003)$ cd ~/a1003
    (a1003)$ ls models/ma16k2401a
    (a1003)$ ls data/eval_clean
    (a1003)$ ls conf/decode_asr.yaml

Edit inference.sh and run it.

    (a1003)$ bash inference.sh

Scoring results

Open results/1best_recog/text

    (a1003)$ python local/uttcer.py data/eval_clean/text results/1best_recog/text

    (a1003)$ sed 's,[.?!,],,g' results/1best_recog/text > results/1best_recog/text.2
    (a1003)$ python local/uttcer.py data/eval_clean/text results/1best_recog/text.2