Skip to content

Latest commit

 

History

History
35 lines (26 loc) · 1.46 KB

README.md

File metadata and controls

35 lines (26 loc) · 1.46 KB
Raw

Simple LJSpeech Dataset Maker

This is a simple LJSpeech Dataset Maker, based on LJSpeechTools. It splits and transcribes the inputs WAV files. Underthehood, it uses Google Speech Recognition for transcriping. Single speaker only.

Usage

  1. Install the dependencies.

    pip install -r requirements.txt

  2. Place your WAV files into the input folder. Mono 22khz WAV is ideal.

  3. Run the pipeline.

    usage: pipeline.py [-h] [-p PARALLEL] [-l SEGMENT_LENGTH] [-s SILENCE_THRESHOLD] [-d DELAY]

LJSpeech Dataset Maker

options:
-h, --help            show this help message and exit
-p PARALLEL, --parallel PARALLEL
                        Number of running process. Default is your core/thread count minus 2.
-l SEGMENT_LENGTH, --segment-length SEGMENT_LENGTH
                        The length of a sengment, in seconds. Default is 12 seconds.
-s SILENCE_THRESHOLD, --silence-threshold SILENCE_THRESHOLD
                        The silence threshold for splitting, in dBFS (negative integer). Default is -40.
-d DELAY, --delay DELAY
                        Add a delay to online transcription, in senconds. Default is 0.1 second.
-u DISCARD_UNDER_SECOND, --discard-under-second DISCARD_UNDER_SECOND
                        Discard any segment under this length, in senconds. Default is 1 second.

  4. The output dataset will be in the dataset folder.