Automatic Lyric Recognition

Transcribe Lyrics with the LibriSpeech ASR Model

We are guided by the steps outlined in https://seongminpark.com/blog/20201222a/ which follows the WSJ recipe.

Once you have Docker Desktop downloaded and running, enter the following terminal commands:

git clone https://github.com/MalcolmMashig/automatic-lyric-recognition.git

cd automatic-lyric-recognition

1. Make sure that all audio files to transcribe are located in audio_dir:

export base_dir=pwd

export audio_dir=$base_dir/audio

2. Prepare the Audio and Kaldi Docker container:

source prep-docker.sh

source download-librispeech-model.sh

source prep-audio.sh

3. Enter the container:

docker exec -it $docker_cid bash

4. Run the decoding process:

source docker-delivery/transcribe.sh

5. Press {CTRL p} then {CTRL q} to exit the container.

6. Copy over the transcriptions from the container to your local machine:

decoded_path=opt/kaldi/egs/wsj/s5/exp/chain_cleaned/tdnn_1d_sp/decode_convert_me_tgsmall/score_20/convert_me.ctm

docker cp $docker_cid:$decoded_path $base_dir/transcriptions.ctm