Audio samples : "Bunched LPCNet2 : Efficient Neural Vocoders Covering Devices from Cloud to Edge"


Paper : Click here

Authors

Sangjun Park (Samsung Research, Samsung Electronics, Republic of Korea)

Kihyun Choo (Samsung Research, Samsung Electronics, Republic of Korea)

Joohyung Lee (Samsung Research, Samsung Electronics, Republic of Korea)

Anton V. Porov (PDMI RAS, Russia)

Konstantin Osipov (PDMI RAS, Russia)

June Sig Sung (Mobile eXperience Business, Samsung Electronics, Republic of Korea)

Click here for other works from Samsung Research TTS Team.


Audio Samples

TTS Samples - Female Speaker

Example 1 : It comes to my mind when it's just about to go out of my mind.
Example 2 : The reserves of electrical energy in all of us can be released only in exceptional circumstances.
Example 3 : The clay contains a particular mineral which helps to neutralize the toxins in the pangi.

Systems Example 1 Example 2 Example 3
Original (24kHz)
Original (16kHz)
B-LPCNet
B-LPCNet2-L
B-LPCNet2-R
B-LPCNet2-S
B-LPCNet2-S16

TTS Samples - Male Speaker

Example 1 : Here are some basic rules to keep in mind when you are invited to dinner by a Western family.
Example 2 : I would want to know all about the animals.
Example 3 : Do you want to see the photographs or do you want the prayer book?

Systems Example 1 Example 2 Example 3
Original (24kHz)
Original (16kHz)
B-LPCNet
B-LPCNet2-L
B-LPCNet2-R
B-LPCNet2-S
B-LPCNet2-S16

Copy-Synthesis Samples - Footprint-Efficiency

Example 1 : You'll feel better if you change your mind.
Example 2 : I told the crew that your background.

Systems ne Example 1 Example 2
Original -
B-LPCNet2-L 128
1
B-LPCNet2-R 128
1
B-LPCNet2-S 128
1

Copy-Synthesis Samples - Computational Efficiency

Example on GRUA units (S=1): How about a fun movie to clear your mind?

na Softmax Single Logistic
384
320
272
224
176
144
Original

Example on Bunch Size (na=384): Unable to find what you asked for.

S Softmax Single Logistic
1
2
3
4
5
Original