2016
VoiceIt Technologies: speech recognition engines
Speech recognition systems for VoiceIt Technologies, built three ways: Siamese neural networks, autoencoder and recurrent neural networks, and Mel-frequency cepstral coefficients.
Three approaches to recognizing a voice
- Siamese neural networks in Python and Java, built with Keras and Caffe, learning to tell whether two recordings come from the same speaker.
- Autoencoder and recurrent neural networks in Java, learning compact representations of speech over time.
- A Java engine built on Mel-frequency cepstral coefficients (MFCC), the classic features for describing how a voice sounds.
Technology
JavaPythonKerasCaffe
Areas
AINeural networksVoice