Skip to content

PyTorch code for end-to-end spoken language understanding (SLU) with ASR-based transfer learning

License

Notifications You must be signed in to change notification settings

VeroJulianaSchmalz/end-to-end-SLU

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

End-to-End Spoken Language Understanding (SLU) in PyTorch

This repo contains Pytorch code for training end-to-end SLU models used in the papers "Speech Model Pre-training for End-to-End Spoken Language Understanding" and "Using Speech Synthesis to Train End-to-End Spoken Language Understanding Models".

If you have any questions about this code or have problems getting it to work, please send me an email at <the email address listed for Loren in the paper>.

Dependencies

PyTorch, torchaudio, numpy, soundfile, pandas, tqdm, textgrid.py

Training

First, change the asr_path and/or slu_path in the config file (like experiments/no_unfreezing.cfg, or whichever experiment you want to run) to point to where the LibriSpeech data and/or Fluent Speech Commands data are stored on your computer.

SLU training: To train the model on an SLU dataset, run the following command:

python main.py --train --config_path=<path to .cfg>

ASR pre-training: Note: the experiment folders in this repo already have a pre-trained LibriSpeech model that you can use. LibriSpeech is pretty big (>100 GB uncompressed), so don't do this part unless you want to re-run the pre-training part with different hyperparameters. If you want to do this, you will first need to download our LibriSpeech alignments here, put them in a folder called "text", and put the LibriSpeech audio in a folder called "audio". To pre-train the model on LibriSpeech, run the following command:

python main.py --pretrain --config_path=<path to .cfg>

Inference

You can perform inference with a trained SLU model as follows (thanks, Nathan Folkman!):

import data
import models
import soundfile as sf
import torch

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
config = data.read_config("experiments/no_unfreezing.cfg"); _,_,_=data.get_SLU_datasets(config)
model = models.Model(config).eval()
model.load_state_dict(torch.load("experiments/no_unfreezing/training/model_state.pth", map_location=device)) # load trained model

signal, _ = sf.read("test.wav")
signal = torch.tensor(signal, device=device).float().unsqueeze(0)

model.decode_intents(signal)

The test.wav file included with this repo has a recording of me saying "Hey computer, could you turn the lights on in the kitchen please?", and so the inferred intent should be {"activate", "lights", "kitchen"}.

Citation

If you find this repo or our Fluent Speech Commands dataset useful, please cite our papers:

  • Loren Lugosch, Mirco Ravanelli, Patrick Ignoto, Vikrant Singh Tomar, and Yoshua Bengio, "Speech Model Pre-training for End-to-End Spoken Language Understanding", Interspeech 2019.
  • Loren Lugosch, Brett Meyer, Derek Nowrouzezahrai, and Mirco Ravanelli, "Using Speech Synthesis to Train End-to-End Spoken Language Understanding Models", ICASSP 2020.

About

PyTorch code for end-to-end spoken language understanding (SLU) with ASR-based transfer learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%