Speech recognition system V1.0
▲ 1. Professional off-line speech recognition model and speech transcription recognition rate: customized public security method speech recognition model must be provided.
2. Must have keyword optimization ability: support user-defined hot words to identify person names and company organization names, and the recognition accuracy of proper nouns reaches 80%; The pre learning mechanism of files and documents for events can better deal with the identification of various fields, and the accuracy is relatively improved by 40%. For different types of events and different regional accents, a customized engine in the field of professional legal affairs is formed through machine learning of a large number of laws, regulations and transcripts, so as to realize the effective recognition of French and French in the process of intelligent recording and broadcasting.
3. It must support seamless access of subsequent capabilities, such as intelligent voice input method and other artificial intelligence technologies.
4. System stability: it can run continuously and stably for 7 * 24 hours.
5. Speech recognition real-time rate: speech recognition real-time rate < 0.3.
6. The system supports a recognition rate of more than 97% for Chinese Putonghua, and more than 60% for Putonghua dialects.
7. Offline recording data recognition: the system must support post recognition and transcribing of recording files.
8. Time index: the transfer result must support the time index function.
9. Professional speech recognition model: it must support the function of customizing special domain speech recognition model.
10. Role separation function: the role separation function must be supported to distinguish the voice and text correspondence of different personnel.
11. Protocol support: it must support WebService, HTTP, socket and other general protocols.
▲ 12. The platform must have extensive software and hardware compatibility, provide standard and open APIs and SDK extensions supporting windows and Linux platforms, and be easy for users to integrate.
12. ★ the speech recognition platform shall meet the real-time voice transfer of 6 interrogation rooms.
★ provide voice optimization service. It can expand the development of local language and dialect recognition.