Frequently Asked Questions

• Audio gallery
  
• Speech processing glossary

1. Can automatic speech recognition be used to transcribe unrestricted broadcast data?

Yes, but the accuracy of speech recognition varies greatly depending upon a wide number of factors, including the type of speech (from prepared to spontaneous speech and conversational speech) and the noise level. So you can expect very good results when transcribing the speech of an anchor speaker in a TV or radio news journal but much less good results for the speech of someone engaged in a very casual conversation.

2. Can automatic transcriptions be used the same way I process text?

Yes, the output of a Vecsys Research LVCSR system is an XML file that can be converted in plain punctuated text by discarding additionnal information such as word timecodes and word confidence scores.

3. I heard that you are developping ASR for many languages. How long it take to develop an ASR for a specific language?

This depends greatly on the available language ressources for the specific language. It also depends on the type of speech data you want process. We are supporting many languages, including Arabic, Dutch, English, French, Mandarin, Russian and Spanish. Contact us to get a more precise answer for the languages you are interested in.

4. Do I need to configure anything to run an LVCSR on broadcast data, like the system vocabulary or the system grammar?

No, the Vecsys Research LVCSR systems come with fully trained language models, so the only information you have to provide to the system is the language being spoken if you know it. It the language is not known, the language can be identified automatically by using a Vecsys Research language recognition software. A language identification system identifies the language being spoken from the speech signal.