Speech-to-Text | Speech & Voice Recognition | ASR

Speech recognition is about converting the human voice into text. You speak, and the technology converts your spoken words into texts.

In a multi-cultural multilingual environment like Hong Kong, people often mix English words with Cantonese or Chinese words when they speak.  However, they speak these English words with unique non-native speaking accents, in contrast to native-speaking accents such as American English or British English.

Everyone has a unique voice.  Even when the same person speaks the same word several times, the acoustic voice signal would be different every time.  Due to co-articulation, the acoustic signal of the word can also be affected by the words before and after it.  Technologists call it a stochastic (or non-deterministic) process, requiring sophisticated probabilistic modeling techniques. Environmental noise presents more hurdles for the technology to overcome.

Though Chinese is a tonal language, speech recognition in Putonghua is not necessarily more challenging than in English. However, Cantonese is more like a verbal language than a written language. The abundance of trendy idiosyncrasy, colloquialism, and slang in spoken Cantonese can sometimes cause difficulties even to local people in Hong Kong.

Speech recognition has always been a multi-disciplinary discipline involving computer science, artificial intelligence, signal processing, information theory, acoustics, language, and linguistics, among others.  With recent technology advances in artificial intelligence (AI), machine learning (ML), and deep neural networks (DNN), speech recognition has become more interdisciplinary than ever before.

Over the past few decades, speech recognition systems in Cantonese, Putonghua, and English have been successfully deployed in a wide variety of industries in Hong Kong and Asia. Successful commercial deployment of speech recognition technology, especially in mission-critical high-traffic applications for public use, requires experience, expertise, and knowhow that are above and beyond simplistic laboratory experiments.

There are other names for speech recognition, such as automatic speech recognition (ASR), speech-to-text (STT), voice recognition (VR), voice-to-text (VTT), 語音辨識, 文語轉換, 語音控制, 語音轉文字.  They all mean more or less the same technology.  From our perspective, Cantonese speech recognition is the same as Cantonese speech to text, and Chinese speech recognition is the same as Chinese speech to text.  However, Hong Kong speech recognition (or Hong Kong speech-to-text) is unique in that people speaking Cantonese often mix it with English.  Thus, speech recognition in Hong Kong (or speech-to-text in Hong Kong) allows the speaker to mix Cantonese with English.  The same applies to Putonghua/Mandarin speech recognition or Putonghua/Mandarin speech-to-text.