A Bit-Stream-Based Modified MELP Vocoder Using Autoencoder

In this paper, we propose a bit-stream-based modified MELP vocoder using autoencoder. This technique trains an autoencoder using the bit-streams extracted from the encoder of an existing vocoder, and the trained autoencoder is divided into an encoder and a decoder model of the modified vocoder. In addition, the bottleneck features of the autoencoder are scalar quantized and transmitted as bit-streams. Through objective speech intelligibility evaluation experiments, we show that the performance of the proposed method is similar to the existing vocoder.