python

Kaldi-Serve Python Binding

Python binding for the kaldiserve C++ library.

Installation

Build from source

You will need pybind11 to be present and built, or alternately you can pass the -DBUILD_PYBIND11=ON flag and cmake will take care of it. You can build the bindings by passing -DBUILD_PYTHON_MODULE=ON -DPYTHON_EXECUTABLE=${which python} options to the main cmake command:

# build the python bindings (starting from current dir)
cd ../build
cmake .. -DBUILD_PYBIND11=ON -DBUILD_PYTHON_MODULE=ON -DPYTHON_EXECUTABLE=${which python}
make -j${nproc}

# copy over the built shared library to the python package
cp python/kaldiserve_pybind*.so ../python/kaldiserve/

# build the python package
cd ../build/python
pip install . -U

Now you can import kaldiserve into your python project.

Docker Image

Using pre-built images

You can pull pre-built docker images (we currently support python version 3.6) from our Docker Hub repository:

docker pull vernacularai/kaldi-serve:latest-py3.6
docker run -it vernacularai/kaldi-serve:latest-py3.6

You will find Python 3.6 pre-installed with kaldiserve python package.

Building the image

You can also build the docker image using the Dockerfile provided (make sure to do it from the root dir):

cd ../
docker build -t kaldi-serve:py${PY_VERSION} -f python/Dockerfile .

Getting Started

Transcription Interface

from io import BytesIO
from kaldiserve import ChainModel, Decoder, parse_model_specs, start_decoding

# model specification
model_spec = parse_model_specs("../resources/model-spec.toml")[0]
# chain model contains all large const components that can be shared across decoders on multiple threads
model = ChainModel(model_spec)

# initialize a decoder that keeps a const reference to the model
decoder = Decoder(model)

audio_files = ["sample1.wav", "sample2.wav"]

for audio_file in audio_files:
    # read audio file as bytes
    with open(audio_file, "rb") as f:
        audio_bytes = BytesIO(f.read()).getvalue()

    with start_decoding(decoder):
        # decode the audio
        decoder.decode_wav_audio(audio_bytes)
        # get the transcripts (10 alternatives)
        alts = decoder.get_decoded_results(10)
    
    print(alts)

Sample Scripts

You will need kaldiserve python package and some other dependencies to be installed:

pip install -r scripts/requirements.txt

There are some sample scripts provided that can be referenced as examples:

1. Transcribe - transcribes a single audio file
2. Batch Transcribe - transcribes a batch of audio files via multi-threading

Known Issues

1. If you face INTEL MKL ERROR when instantiating ChainModel|DecoderFactory|DecoderQueue, try the following:

export LD_PRELOAD="${MKL_ROOT}/lib/intel64/libmkl_rt.so"

2. If you face *** Error in python: double free or corruption (!prev): ..., try the following:

apt-get install libtcmalloc-minimal4
export LD_PRELOAD="/usr/lib/libtcmalloc_minimal.so.4"