Hi Saniya, I have just come across your question. I appreciate this may no longer be timely but I am adding a couple pointers in case they can help others.
For isolating or segmenting speech in low-noise recordings, the function detectSpeech should work just fine. This will return start and stop times of all signal regions where speech is detected, but no text "transcription" of the actual speech content.
To estimate the transcription you will need a speech-to-text model based on machine learning. The following two links will be relevant:
- speech2text, also availale from within Signal Labeler per this example. Note that the use of this function also requires a registration with a cloud-based speech-to-text service from either Google, Microsoft, or IBM. Refer to the documentation for details
- The MATLAB implementation of the wav2vec 2.0 deep learning network, available from here on GitHub. This will only work for English but it is completely based on MATLAB
I hope this helps.
