Wake word detection modeling for Firefox Voice, supporting open datasets like Google Speech Commands and Mozilla Common Voice.
Citation:
@inproceedings{tang-etal-2020-howl,
title = "Howl: A Deployed, Open-Source Wake Word Detection System",
author = "Tang, Raphael and Lee, Jaejun and Razi, Afsaneh and Cambre, Julia and Bicking, Ian and Kaye, Jofish and Lin, Jimmy",
booktitle = "Proceedings of Second Workshop for NLP Open Source Software (NLP-OSS)",
month = nov,
year = "2020",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.nlposs-1.9",
doi = "10.18653/v1/2020.nlposs-1.9",
pages = "61--65"
}
-
git clone https://github.com/castorini/howl && cd howl -
Install PyTorch by following your platform-specific instructions.
-
Install PyAudio and its dependencies through your distribution's package system.
-
pip install -r requirements.txt -r requirements_training.txt(some apt packages might need to be installed) -
./download_mfa.shto setup montreal forced aligner (MFA) for dataset generation
Generating a dataset for a custom wakeword requires three steps:
- Generating raw audio dataset that howl can load from open datasets
- Generate orthographic transcription alignments for each audio file.
- Attach the alignment to the raw audio dataset generated in step 1.
Having said that we recommend Common Voice dataset for the open audio datasets and Montreal Forced Aligner (MFA) for the transcription alignment.
Downloading MFA can be achieved simply running download_mfa.sh script. Along with the aligner, the script will download necessary English pronunciation dictionary.
Once they are ready, a dataset can be generated using the following script.
./generate_dataset.sh <common voice dataset path> <underscore separated wakeword (e.g. hey_fire_fox)> <inference sequence (e.g. [0,1,2])> <(Optional) "true" to skip negative dataset generation>For detailed explanation, please refer to How to generate a dataset for custom wakeword
- Source the relevant environment variables for training the
res8model:source envs/res8.env. - Train the model:
python -m training.run.train -i datasets/fire/positive datasets/fire/negative --model res8 --workspace workspaces/fire-res8. It's recommended to also use--use-stitched-datasetsif the training datasets are small. - For the CLI demo, run
python -m training.run.demo --model res8 --workspace workspaces/fire-res8.
train_model.sh is also available which encaspulates individual command into a single bash script
./train_model.sh <env file path (e.g. envs/res8.env)> <model type (e.g. res8)> <workspace path (e.g. workspaces/fire-res8)> <dataset1 (e.g. datasets/fire-positive)> <dataset2 (e.g. datasets/fire-negative)> ...howl-models contains workspaces with pretrained models
To get the latest models, simply run git submodule update --init --recursive
VOCAB='["hey","fire","fox"]' INFERENCE_SEQUENCE=[0,1,2] INFERENCE_THRESHOLD=0 NUM_MELS=40 MAX_WINDOW_SIZE_SECONDS=0.5 python -m training.run.demo --model res8 --workspace howl-models/howl/hey-fire-fox-
Install PyAudio and PyTorch 1.5+ through your distribution's package system.
-
Install Howl using
pip
pip install howl
- To immediately use a pre-trained Howl model for inference, we provide the
clientAPI. The following example (also found underexamples/hey_fire_fox.py) loads the "hey_fire_fox" pretrained model with a simple callback and starts the inference client.
from howl.client import HowlClient
def hello_callback(detected_words):
print("Detected: {}".format(detected_words))
client = HowlClient()
client.from_pretrained("hey_fire_fox", force_reload=False)
client.add_listener(hello_callback)
client.start().join()
First, follow the installation instructions in the quickstart guide.
- Download the Google Speech Commands dataset and extract it.
- Source the appropriate environment variables:
source envs/res8.env - Set the dataset path to the root folder of the Speech Commands dataset:
export DATASET_PATH=/path/to/dataset - Train the
res8model:NUM_EPOCHS=20 MAX_WINDOW_SIZE_SECONDS=1 VOCAB='["yes","no","up","down","left","right","on","off","stop","go"]' BATCH_SIZE=64 LR_DECAY=0.8 LEARNING_RATE=0.01 python -m training.run.pretrain_gsc --model res8
- Download the Hey Firefox corpus, licensed under CC0, and extract it.
- Download our noise dataset, built from Microsoft SNSD and MUSAN, and extract it.
- Source the appropriate environment variables:
source envs/res8.env - Set the noise dataset path to the root folder:
export NOISE_DATASET_PATH=/path/to/snsd - Set the firefox dataset path to the root folder:
export DATASET_PATH=/path/to/hey_firefox - Train the model:
LR_DECAY=0.98 VOCAB='["hey","fire","fox"]' USE_NOISE_DATASET=True BATCH_SIZE=16 INFERENCE_THRESHOLD=0 NUM_EPOCHS=300 NUM_MELS=40 INFERENCE_SEQUENCE=[0,1,2] MAX_WINDOW_SIZE_SECONDS=0.5 python -m training.run.train --model res8 --workspace workspaces/hey-ff-res8
- Download hey snips dataset
- Process the dataset to a format howl can load
VOCAB='["hey","snips"]' INFERENCE_SEQUENCE=[0,1] DATASET_PATH=datasets/hey-snips python -m training.run.deprecated.create_raw_dataset --dataset-type 'hey-snips' -i ~/path/to/hey_snips_dataset- Generate some mock alignment for the dataset, where we don't care about alignment:
python -m training.run.attach_alignment \
--input-raw-audio-dataset datasets/hey-snips \
--token-type word \
--alignment-type stub- Use MFA to generate alignment for the dataset set:
mfa_align datasets/hey-snips/audio eng.dict pretrained_models/english.zip datasets/hey-snips/alignments- Attach the MFA alignment to the dataset:
python -m training.run.attach_alignment \
--input-raw-audio-dataset datasets/hey-snips \
--token-type word \
--alignment-type mfa \
--alignments-path datasets/hey-snips/alignments- Source the appropriate environment variables:
source envs/res8.env - Set the noise dataset path to the root folder:
export NOISE_DATASET_PATH=/path/to/snsd - Set the noise dataset path to the root folder:
export DATASET_PATH=/path/to/hey-snips - Train the model:
LR_DECAY=0.98 VOCAB='["hey","snips"]' USE_NOISE_DATASET=True BATCH_SIZE=16 INFERENCE_THRESHOLD=0 NUM_EPOCHS=300 NUM_MELS=40 INFERENCE_SEQUENCE=[0,1] MAX_WINDOW_SIZE_SECONDS=0.5 python -m training.run.train --model res8 --workspace workspaces/hey-snips-res8
howl also provides a script for transforming howl dataset to mycroft-precise dataset
VOCAB='["hey","fire","fox"]' INFERENCE_SEQUENCE=[0,1,2] python -m training.run.generate_precise_dataset --dataset-path /path/to/howl_datasetTo verify the correctness of our implementation, we first train and evaluate our models on the Google Speech Commands dataset, for which there exists many known results. Next, we curate a wake word detection datasets and report our resulting model quality.
For both experiments, we generate reports in excel format. experiments folder includes sample outputs from the for each experiment and corresponding workspaces can be found here
For command recognition, we train the four different models (res8, LSTM, LAS encoder, MobileNetv2) to detect twelve different keywords: “yes”, “no”, “up”, “down”, “left”, “right”, “on”, “off”, “stop”, “go”, unknown, or silence.
python -m training.run.eval_commands_recognition --num_iterations n --dataset_path < path_to_gsc_datasets >In this experiment, we train our best commands recognition model, res8, for hey firefox and hey snips and evaluate them with different threashold.
Two different performance reports are generated, one with the clean audio and one with audios with noise
python -m training.run.eval_wake_word_detection --num_models n --hop_size < number between 0 and 1 > --exp_type < hey_firefox | hey_snips > --dataset_path "x" --noiseset_path "y"We also provide a script for generating ROC curve. exp_timestamp can be found from the reports generated from previous command
python -m training.run.generate_roc --exp_timestamp < experiment timestamp > --exp_type < hey_firefox | hey_snips >
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent