A slim Python client for Mozilla's DeepSpeech speech-to-text
from src.deepspeech_stt import deepspeech_predict
ouput_text: str = deepspeech_predict(
wav_file_path,
batch_after_silence=True,
silence_threshold=45, # 45db
filters=["logmmse_denoise", "butter_bandpass_filter"]
)| Parameter | Default | Description |
|---|---|---|
wave_filename |
None |
Path to wave file |
batch_after_silence |
True |
Create batch from input splitting after natural gaps of silence |
silence_threshold |
50 |
The threshold (in decibels) below reference to consider as silence |
filters |
None |
List of signal filters to apply as pre-processing:butter_bandpass_filter, high_pass_filter, low_pass_filter, logmmse_denoise |
| See notebook for examples |
Download Mozilla's DeepSpeech 0.7.4 pre-trained model (~200mb)
Then run:
poetry install
poetry shell