Persian Kaldi Profile

A Rhasspy profile for Persian (fa).

Trained from approximately 293 hours of audio from Common Voice (Persian 7.0 dataset, validated, 10% test).

Available Vosk models:

Small nnet3
- WER: 15.57%
Large nnet3
- WER: 13.58%

Installation

Get started by first installing Vosk:

# Create virtual environment
python3 -m venv .venv
source .venv/bin/activate
pip3 install --upgrade pip
pip3 install --upgrade wheel setuptools

# Install Vosk
pip3 install vosk

Next, download the model and extract it:

wget 'https://github.com/rhasspy/fa_kaldi-rhasspy/releases/download/v1.0/vosk-model-small-fa-rhasspy-0.15.zip'
unzip vosk-model-small-fa-rhasspy-0.15.zip

Finally, run the transcribe.py Python program with the model and an audio file:

python3 transcribe.py vosk-model-small-fa-rhasspy-0.15 welcome.wav

{"result": [{"conf": 1.0, "end": 0.48, "start": 0.06, "word": "خوش"}, {"conf": 1.0, "end": 1.11, "start": 0.48, "word": "آمدید"}], "text": "خوش آمدید"}

For each audio file given to transcribe.py, a line of JSON will be printed in the output with the transcription details.

Troubleshooting

if you face this error:

ModuleNotFoundError: No module named "numpy"

check this issue

if you face this error:

ModuleNotFoundError: No module named "librosa"

check this issue

if you face this error:

/your/path/.venv/lib/python3.9/site-packages/librosa/util/decorators.py:88: UserWarning: PySoundFile failed. Trying audioread instead.
  return f(*args, **kwargs)

check this issue

If you got any other error feel free to open an new issue here and describe your problem specific.

GameO7er/fa_kaldi-rhasspy

Persian Kaldi Profile

Installation

Troubleshooting