speech-enhancement

DTLN Speech Enhancement sample

The open source model DTLN was taken from the github repository in the link.

Setup

Download the model file

Clone the model-interfaces repository and navigate to the speech-enhancement reference example:

git clone [email protected]:aixplain/aixplain-models-internal.git
cd aixplain-models-internal/docs/user/samples/speech-enhancement

Download the model file using the commands

wget https://aixplain-kserve-models-dev.s3.amazonaws.com/serving-models/sample-models/speech-enhancement/dtln/dtln/saved_model.pb
wget https://aixplain-kserve-models-dev.s3.amazonaws.com/serving-models/sample-models/speech-enhancement/dtln/dtln/variables/variables.data-00000-of-00001
wget https://aixplain-kserve-models-dev.s3.amazonaws.com/serving-models/sample-models/speech-enhancement/dtln/dtln/variables/variables.index

Place it in a folder called dtln in the current directory. The files should be placed in the following tree's structure:

dtln
| - saved_model.pb
| - variables
    | - variables.data-00000-of-00001
    | - variables.index

Install requirements

sudo apt-get install ffmpeg

# Install model-interfaces from GitHub, preferably by using a virtualenv
pip install -e 'git+https://[email protected]/aixplain/aixplain-models-internal.git@master#egg=model_interfaces'

pip install -r src/requirements.txt

• GH_ACCESS_TOKEN: Generate a GitHub access token from your account that can clone the model_interfaces repository

Documentation to generate the GitHub personal access token can be found here.

Run the model locally

Test using the given test and samples

ASSET_DIR=. ASSET_URI=dtln pytest

Serve the model using a webserver

ASSET_DIR=. ASSET_URI=dtln python src/model.py

Test the model server

python src/sample_request.py

Build and Run the model with docker

Navigate to the speech-enhancement reference example:

cd docs/user/samples/speech-enhancement

Build the model's container

docker build --build-arg GH_ACCESS_TOKEN=<TOKEN_FROM_GITHUB_ACCOUNT> --build-arg ASSET_URI=<ASSET_URI> . -t 535945872701.dkr.ecr.us-east-1.amazonaws.com/aixmodel-dtln

• GH_ACCESS_TOKEN: Generate a GitHub access token from your account that can clone the model_interfaces repository
• ASSET_URI: Your model's name; 'dtln' in this reference example.

Run the container

docker run -e ASSET_DIR=/ -e ASSET_URI=dtln -p 8080:8080 535945872701.dkr.ecr.us-east-1.amazonaws.com/aixmodel-dtln