Fix typo and calculation code in Japanese detection.

[miiton]

Hi.

I think it's probably wrong, so I've fixed the following:

• mandrin to mandarin
• Does not include katakana in total

[greyblake]

Quality benchmarks:

Before:

| LANG     | AVG    | <= 20  | 21-50  | 51-100 | > 100  |
|----------|--------|--------|--------|--------|--------|
| Mandarin | 96.08% | 97.54% | 96.92% | 95.45% | 94.43% |
| Japanese | 94.38% | 93.97% | 96.04% | 94.31% | 93.18% |
| AVG      | 95.23% | 95.76% | 96.48% | 94.88% | 93.80% |

OVERALL: 2 languages
AVG: 95.23%

After

| LANG     | AVG    | <= 20  | 21-50  | 51-100 | > 100  |
|----------|--------|--------|--------|--------|--------|
| Mandarin | 96.09% | 97.54% | 96.92% | 95.45% | 94.43% |
| Japanese | 94.37% | 93.97% | 96.04% | 94.30% | 93.18% |
| AVG      | 95.23% | 95.76% | 96.48% | 94.88% | 93.81% |

OVERALL: 2 languages
AVG: 95.23%

It's slightly (0.01%) improved correct detection rate of Cmn and decreased correct detection rate of Japanese. On average it remains the same.

[greyblake]

@miiton Thanks!