In a ASR project, a boring part is the creation of a speech dataset. This script makes the dictation of a set of transcriptions easier. It was initially developed as a tool in my GSoC project.
pip install sounddevice
We assume that there is a folder that contains a set of texts for dictation. At first, it displays the next text to be dictated, it waits for some seconds and then it starts the recording. When the user finishes the curent recording, he presses Ctrl+C and the file is saved automatically as a .wav file.
$ python recorder.py -h
usage: recorder.py [-h] --output OUTPUT --input INPUT --rate RATE --channels
CHANNELS [--delay DELAY]
Tool for recording a speech dataset quickly.
optional arguments:
-h, --help show this help message and exit
required arguments:
--output OUTPUT Folder to save the files
--input INPUT Folder that contains the input transcriptions
--rate RATE Sample rate to be used
--channels CHANNELS Number of channels to be used
optional_arguments:
--delay DELAY Delay until recording starts
For example, if we have two transcriptions in data folder, in order to dictate them in Sphinx format we do the following:
$ python recorder.py --output wav --input data/ --rate 16000 --channels 1 --delay 2
press n to record data/data_0 or q to quit
n
Text to dictate:
test1
Recording...
press Ctrl+C to stop the recording
^C
Recording finished: 'wav/data_0.wav'
press n to record data/data_1 or q to quit
n
Text to dictate:
test2
Recording...
press Ctrl+C to stop the recording
^C
Recording finished: 'wav/data_1.wav'
This tool is licensed under the MIT License.