Logo of Cloudcraftz

Artificial Intelligence in Multi Lingual Voice Analytics 

In the hyper competitive world of banking, it is critical to drive up selling of products and increasing customer engagement using voice chats in the language retail customers are most comfortable. Due to the customer comfort, majority of such conversations happen to be a mix of English banking terms embedded within the vernacular language voice chat. At the same time, it very important that voice chat outreach follows laid down banking and regulatory guidelines, e.g. following certain etiquettes, not asking customers to share confidential information like password, mobile otp or avoid mis-selling of products by exaggerating product features or not covering certain aspects.  

The current solution was to carry out random sample-based audits of the conversation which could lead of lower rate of compliance or is not easy to scale from cost and people perspective.  

The solution uses a custom trained speech to text models for converting voice chats to text followed by a machine learning model to monitor all voice chats with exception-based tagging problematic chats for review and intervention.  

Innovative Approach and Methodology

The solution made use of multi model speech to text models using custom trained model specific to the customer conversations combined with other commercial speech to text models in language like English, Hindi, Bengali and Tamil. Instead of going for a faithful multilingual transcript of the chat, the approach used the above multi-state model pipeline towards identification and analysis of the relevant fragments of chat responsible for client’s compliance guidelines. The architecture provides ability to configure the compliance policies on the fly with ease without changes to the code or the model weights.  

The solution continues to evolve as Speech-to-Text models become more power and multi modal. The ability to custom train the model for specific client data with newer fine-tuning approaches also continues to add more value to the solution.  

Key Features and Capabilities

  • Multi-model speech-to-text pipeline: Combines custom-trained models specific to customer conversations with commercial speech-to-text models in multiple languages for accurate analysis of relevant chat fragments. 
  • Configurable compliance policies: Allows easy configuration of compliance policies without code or model weight changes, adapting to clients’ specific needs and regulatory requirements. 
  • Exception-based reporting: Provides on-demand, exception-based reporting, highlighting problematic chats for efficient monitoring and prompt intervention. 

Conclusion

The Artificial Intelligence in Multi Lingual Voice Analytics solution showcases an innovative approach to leveraging custom-trained speech-to-text models for analytics and driving customer engagement in voice chats within the banking field. By combining custom-trained models specific to customer conversations with commercial speech-to-text models in various languages, the solution accurately identifies and analyzes relevant chat fragments related to the client’s compliance guidelines. 

The configurable compliance policies and exception-based reporting features enable clients to easily adapt the solution to their specific needs and efficiently monitor potential compliance issues. As speech-to-text models continue to evolve, the solution will further enhance its capabilities, making it an essential tool for driving customer engagement and ensuring compliance in the highly competitive banking industry. 

Get in Touch

Get in Touch

Fill up the form and we'll get back to you within two business days





    captcha