CLTranscriptor 0.0.6

Wav2Vec2-based transcriptor fine tuned on chilean lessons

CLTranscriptor

Wrapper for spanish speech-to-text model based on huggingface's Wav2Vec2ForCTC and finetuned on Chilean lessons + PySpellChecker's spanish spellchecking algorithm.

Install

To install, simply use pip:

pip install cltranscriptor

Usage

To use, initialize a Transcriptor object:

from cltranscriptor.cltranscriptor import Transcriptor
transcriptor = Transcriptor()

By default, spell checking is set to True and the model name is the one available at dannersm/wav2vec2-large-xlsr-53-chilean-lessons, which is based on Jonatas Grosman's model and finetuned on a 6 hour set of chilean lessons.

To transcribe a file, call Transcriptor.transcribe():

transcriptor.transcribe('/path/to/your/audio_file.wav')

By default, the file is streamed into 10 second intervals (to avoid loading it in memory) and returns a list with the transcripts for each segment. If you want to transcribe a relatively short file all at once you can pass interval=None:

transcriptor.transcribe('my_file.wav', interval=None)

You can also pass the offset and duration parameters which will be passed to librosa.stream to set the start time and a maximum duration to the transcription

transcriptor.transcribe('my_file.wav', offset=600, duration=120) # transcribe 2 minutes of audio starting from minute 10

Finally, you can control the length of the streamed segments passing interval:

transcriptor.transcribe('my_file.wav', interval=15) # transcribe every 15 seconds