Team Pepe solution for Openvino hackathon: emotions and speaker recognition from voice

To use your own voice in speaker recognition model, please record it with following command:

python3 voice_db/record_voice.py %filename%.wav

Recording should last about 10-15 seconds of continuous speech. Press Ctrl+C to interrupt the recording process. After that you should generate a spectrograms for all audio files. That could be done with:

python3 create_base.py

It will convert all recorded voices to .npy files with spectrograms.

python3 demo.py

Note that speech sample should lasts about 10 seconds, otherwise it will not be recognized (due to speaker recognition model input_shape) It will output list with all emotions probabilities and cosine distances for all speech samples which were stored in voice_db folder. In our demo we used a threshold (=0.35) to identificate a speaker.

Speaker recognition: https://github.com/linhdvu14/vggvox-speaker-identification
Emotions recognition: https://github.com/alexmuhr/Voice_Emotion