/blossom

Blossom is a library for Wake-up word detection problem.

Primary LanguagePythonMIT LicenseMIT

Table of contents

  1. Introduction

  2. How to use Blossom

  3. Reference

Blossom

Blossom is a library for Wake-up word detection problem use Multihead attention RNN (MHAtt-RNN) architecture.

drawing

How to use Blossom

Installation 

>>> python setup.py bdist_wheel

>>> pip install dist/blossom-0.0.1-py3-none-any.whl

Data Structure

data
    trigger
        train
            right
                right_1.wav
                right_2.wav
                ...
            
            on
                on_1.wav
                on_2.wav
                ...

            ...

        valid

        test

Example usage

import os

from blossom.models import MHAttKWS
from blossom.datasets import MHAttDataset
from blossom.learners import MHAttKWSLearner

def test_train():

    train_dataset = MHAttDataset(
        mode='train',
        root='./data/trigger'
    )
    valid_dataset = MHAttDataset(
        mode='valid',
        root='./data/trigger'
    )
    test_dataset = MHAttDataset(
        mode='test',
        root='./data/trigger'
    )

    model = MHAttKWS(
        num_classes=2,
        in_channel=1,
        hidden_dim=128,
        n_head=4,
        dropout=0.1
    )

    learner = MHAttKWSLearner(model=model)
    learner.train(
        train_dataset=train_dataset,
        test_dataset=test_dataset,
        batch_size=48,
        learning_rate=1e-4,
        eps=1e-8,
        betas=(0.9, 0.999),
        max_steps=5,
        n_epochs=1,
        shuffle=True,
        num_workers=8,
        view_model=True,
        save_path='./models',
        model_name='mhatt_model'
    )
    
test_train()


def test_inference():
    model = MHAttKWS(
        num_classes=2,
        in_channel=1,
        hidden_dim=128,
        n_head=4,
        dropout=0.1
    )

    learner = MHAttKWSLearner(model=model)
    learner.load_model(model_path='./models/mhatt_model.pt')

    output = learner.inference(input='data/trigger/test/on/5c8af87a_nohash_3.wav')

    print(output)


test_inference()

Reference

  1. Douglas Coimbra de Andrade, Sabato Leo, Martin Loesener Da Silva Viana and Christoph Bernkopf: “A neural attention model for speech command recognition”, in arXiv:1808.08929, 2018.

  2. Oleg Rybakov, Natasha Kononenko, Niranjan Subrahmanya, Mirko Visontai and Stella Laurenzo: “Streaming keyword spotting on mobile devices”, in arXiv:2005.06720, 2020.

Author

(To request sample data for experiment. Please contact by my email. )