The Large-Vocabulary Speech Recognition engine provides full speech transcription capability on small embedded devices as well as servers. The speech recognition engine supports a configurable lexicon, in the order of 300,000 unique words, with the capability of customizing a language model to your domain of interest. The engine comes with a generic language model which covers most idiosyncrasies of the language, as default. Medical, legal, financial, and other language domains are also available. In addition, new and unique domains may be defined and trained in a matter of hours.

Once utilized in conjunction with our Speaker Recognition engine, it provides full diarization capabilities, where the Speaker Recognition engine segments speakers and labels their identities and the Speech Recognition engine transcribes the text that is spoken by each individual. These engines work together and provide timestamps and other details such as score and confidence for each result. They also provide multiple possible results sorted according to their relevance scores. These results may be returned in XML, JSON, or even clean human readable Text and HTML. We provide a C++ API as well as web, Android, iOS, and command-line interfaces.

Supported Languages

The RecoMadeEasy^® Speech Recognition engine is currently available for the following languages:

English


• All dialects of English
• Supports 8kHz and 16kHz Audio

Spanish


• Major dialects of Spanish
• Supports 8kHz and 16kHz Audio

Chinese (Mandarin)


• Major dialects of Mandarin Chinese
• Supports 8kHz and 16kHz Audio

Arabic


• Major dialects of Arabic
• Supports 8kHz and 16kHz Audio

German


• Major dialects of German
• Supports 8kHz and 16kHz Audio