SPEECH-TO-TEXT TECHNOLOGY FROM СDC
Speech-to-text is a speech recognition technology developed on the basis of modern machine learning methods.
The technology allows converting audio stream into text in real time.
Main characteristics
- Speaker Independent
- Recognition accuracy - up to 84%
- Customizing for a given topic
- Offline work
- Dictionary - 500 thousand words
SCENARIOS OF USE
Dictation of text when there is no possibility to type (in production, transport, etc.)
SCENARIOS OF USE
Automation of meeting and meeting records management
SCENARIOS OF USE
Translation into text of recorded lectures, conferences, etc.
SCENARIOS OF USE
Automation of call centers
SPEAKER RECOGNITION
The service allows for three-factor authentication
- registered phone availability
- voice match
- sync with the text sent
It is based on an original technology that allows
- reduce the voice sample size to five seconds.
- surpass a similar solution from an international company on a standard MOBIO Database data set
- run everything on the CPU
AUDIO RECOGNITION
A mobile device or a low-power device can recognize a set of predefined sound events.
An audio event identification module has been implemented:
- baby crying
- dog barking
- broken glass
- gunshot
- fire siren
Possible use:
- Security and alarm systems
- Implementation of a specified scenario based on an audio event (switching on video cameras, switching on lights, sending alerts, initializing diagnostics of equipment status)
Exactly:
- Child crying - 96%
- Dog barking - 94%
- Glass break - 92%
- Sound of fire/smoke detector - 98%
Diarization
.