At NCSOFT, we studied "broadcasting-style" TTS system that can imitate sportcasters. To express the diverse speaking style of baseball broadcasting, especially in pitch ranges, expressive speech synthesis technique was needed. We studied TTS system that can generate speech with various emotions and its intensity. Furthermore, we also developed the TTS system to generate speech with proper prosody corresponding to text symbols (,, ~, !, ?).
The speech samples generated from neutral-style TTS system and our broadcasting-style TTS system can be found below. The content of the speech samples are same.
Neutral style |
|
Broadcasting style |
Our TTS system can also control the intensity of the epxressivness depend on the situation. Please listen to the speech smaples below.
Weak |
|
Strong |
Furthermore, various situations are bound to appear in baseball broadcasts. The following videos are samples with synthesized speech from our TTS system of four situations: 1) player introduction, 2) commentary on ball counts or pitches, and 3) events such as strike outs, hits, and home runs.
1) Player introduction |
|
2) Commentary on ball counts or pitches |
|
3) Events such as strike outs, hits, and home runs |
|
Lastly, we were also able to create a demo where emotions could be adjusted to provide biased broadcasting for each team in the same situation. Through this, we present a baseball broadcast scenario tailored to each individual's cheering team.
NC-side broadcasting (depressed) |
LG-side braodcasting (excited) |