This repository includes couple options for transcription;
- YouTube video transcriber
- A video transcriber
- Record your voice and transcriber afterwards
- Transcribe while speaking
For each use case's you need to install couple different softwares/libraries.
The main use case is downloading a video / videos (as audio) from YouTube and transcribe them. For this use case, these steps should be followed;
-
clone the repository
$ git clone https://github.com/erolrecep/Transcriber.git -
Install SoX Swiss army knife for audio processing things
$ brew install sox # For Mac Os X $ sudo apt install sox # For Ubuntu -
create a new Python virtual environment
$ conda create --name transcriber python=3.6 $ conda activate transcriber $ conda install tensorflow==1.13.1 # if you have GPU, then install *conda install tensorflow-gpu==1.13.1* (surprisingly I like this version of tensorflow :) ) $ pip install youtube-dl deepspeech==0.7.4 # if you have GPU, then install *pip install deepspeech-gpu==0.7.4* -
Download pre-trained DeepSpeech models from here
- This repository uses 0.7.4 version of the DeepSpeech, you can try the same setup with newer models.
- You need to download 0.7.4 pdmm file
- If you want, you can also download and load scorer provided by Mozilla, scorer
-
Now, the virtual environment is ready, the next step is running the project. For your convenience, I provided a sample .wav file so you can test your setup if it's working. Also, you can download audio files from here
$ python run.py # This will read audio files from the *audio_locations.txt* file. $ python run.py -a audio_files/sample.wav # This will only run inference on this input .wav file