Automatic Language and Speaker Identification from Conversational speech in North East Indian languages

The solution of Automated Language and Speaker Identification system from conversational speech data in North-East Indian languages is deployed at the NE based security agency premises. Solution includes modules for following three major components.

1. Speaker Segmentation: segmenting, separating silence, non-speech and speaker specific speech parts from conversational speech
2. Language Identification (LID): Identifies the language(s) spoken in the speaker specific speech parts within already trained 14 NE Indian languages (Angami, Ao, Assamese, Bodo, Garo, Khasi, Konyak, Kuki, Lotha, Manipuri, Mizo, Nagamese, Sumi, Thankul) spoken in North East states.
3. Automatic Speaker Identification (ASI): Enrolls new speaker and Identifies (label) the Speaker on availability of enrolled speakers’ voice profiles.

Use Cases

• Surveillance
• Information retrieval
• Immigration
• Forensic investigations 

Salient Features

• Individual processing for spoken language and speaker Identification
• Works for two speakers’ conversations and also on single speaker's speech (without Speaker Segmentation)
• Performs on speech data of different environments, medium noise level
• Tested on real time conversation data from NE based security agency
• Cross language matching: Authenticate speakers speaking in different languages