Experiencing Text to Speech
- 主题发起人 xujiajin
- 时间
Speech Recognition Technology Used in CALL Applications in China
Technology in use
We have made a general evaluation of the recognition engines used in the learning software, for example, TeLL me More Pro, LiveABC Interactive (developed by the same company of CNN Interactive), and many others.
Most of the CALL programs are PC-based and some include web-based recognition.
Record/playback/compare method is in the main. During the playback and comparison, pass-fail mode is used to control the learning pace, but the problem is that the technology is unreliable per se. In some cases, individual word waveform evaluation score and score for the general performance are displayed. Others apply the “pass--try again--listen to the original sound--skip” to guide the learning path.
Task design in CALL applications with speech recognition technology
Due to the apparent limitations of the speech processing technology, current voice-interactive CALL applications only allows for the training of lower-level repetitive skills in language learning. But this does not keep us from making up some original tasks or activities with speech recognition technology. Apart from record, playback and comparison, we can also write some good voice-based exercises like (1) oral cloze, in which learners spell out the correct answers from the given options or distracters; (2) situated multiple choice, in which the learners answer the questions orally, that is to choose the right answer from among three or four options. Usually the first question in such an activity is an open one to select a situation for the episode to develop. Different choices anticipate different set of questions. So learners are positively motivated in doing such an exercise and their interests in learning are raised; and (3) role play, which is a popular activity employed by almost all the CALL applications we have reviewed thus far. In a voice-based role play, the learners first select a role of the dialog. Then computer screens out the recording of the role in the dialog the learner chooses and plays the recording of the other role. The learner reads or recites the sentences that the role s/he plays. After either of the roles reads the last sentence, the computer mixes the learner’s and computer’s “talk” and plays it back. The learner can sit back to listen to the dialog s/he has done with the computer. S/he may also choose to listen to the original sound.
Technology in use
We have made a general evaluation of the recognition engines used in the learning software, for example, TeLL me More Pro, LiveABC Interactive (developed by the same company of CNN Interactive), and many others.
Most of the CALL programs are PC-based and some include web-based recognition.
Record/playback/compare method is in the main. During the playback and comparison, pass-fail mode is used to control the learning pace, but the problem is that the technology is unreliable per se. In some cases, individual word waveform evaluation score and score for the general performance are displayed. Others apply the “pass--try again--listen to the original sound--skip” to guide the learning path.
Task design in CALL applications with speech recognition technology
Due to the apparent limitations of the speech processing technology, current voice-interactive CALL applications only allows for the training of lower-level repetitive skills in language learning. But this does not keep us from making up some original tasks or activities with speech recognition technology. Apart from record, playback and comparison, we can also write some good voice-based exercises like (1) oral cloze, in which learners spell out the correct answers from the given options or distracters; (2) situated multiple choice, in which the learners answer the questions orally, that is to choose the right answer from among three or four options. Usually the first question in such an activity is an open one to select a situation for the episode to develop. Different choices anticipate different set of questions. So learners are positively motivated in doing such an exercise and their interests in learning are raised; and (3) role play, which is a popular activity employed by almost all the CALL applications we have reviewed thus far. In a voice-based role play, the learners first select a role of the dialog. Then computer screens out the recording of the role in the dialog the learner chooses and plays the recording of the other role. The learner reads or recites the sentences that the role s/he plays. After either of the roles reads the last sentence, the computer mixes the learner’s and computer’s “talk” and plays it back. The learner can sit back to listen to the dialog s/he has done with the computer. S/he may also choose to listen to the original sound.
[本贴已被 作者 于 2005年07月01日 09时36分11秒 编辑过]
michelhere
初级会员
讯飞的大的语音合成系统都是采用拼接的技术，所以自然度很好，现在的qq汽车上也用了讯飞的语音合成技术，不过汽车上用的是共振峰合成，语言自然度就差了很多，机器味很重。
在At&T的网站上，Bell Lab的网站上都有他们做的英语语音合成，也比较有意思。
在At&T的网站上，Bell Lab的网站上都有他们做的英语语音合成，也比较有意思。
