人工智能語音技術試水網絡鑑黃

An artificiAl intelligence (AI) voice recognition system that can monitor and identify pornographic content began its open beta from last Sunday.

上週日，一款由阿里巴巴公司開發的能監測和識別色情內容的人工智能語音識別系統開始公測。

Helped by the voiceprint recognition method, the Alibaba-developed voice recognition system can identify multiple languages such as Chinese, Japanese, English and Russian, as well as Chinese dialects from different provinces such as Hunan, Hubei, Henan, Sichuan and Guangdong.

藉助聲紋識別技術，這個語音識別系統能夠識別中文、日文、英文、俄文等多國語言，還能識別湖南、湖北、河南、四川、廣東等國內多個省份的方言。

Transforming voice into script, the system compares the scripts with keywords in its lexicon and anti-spam audio models to determine if something is pornographic.

該系統將語音識別轉成文字後，再將這些文字與其詞庫中的關鍵詞或反垃圾音頻模型比對，判斷是否涉黃。

The lexicon and anti-spam audio models collect tens of thousands of pornographic words with the same or similar pronunciations, according to Alibaba.

據阿里巴巴介紹，詞庫和反垃圾音頻模型收集了成千上萬個發音相同或類似的涉黃詞彙。

The system monitors both online and offline voice files. It also has the ability to adapt and "learn" through constant use. For example, its Cantonese recognition ability was cultivated by watching TV series.

款系統可以監控在線和離線的語音文件。此外，它還具有通過持續的使用適應“學習”的能力，比如，其廣東話識別能力是通過看電視劇訓練的。

The system is scheduled to be put into operation in September this year.

預計該系統將於今年9月投入使用。