In celebration of the company's 16th anniversary, Reverie Language Technologies has launched a new speech-to-text (STT) model designed to address India's diverse linguistic environment. This model not only recognizes Hindi and English, but also handles Hinglish and other mixed languages, greatly meeting the needs of industries such as banking and call centers.

Voice Control

Image source note: The image is AI-generated, provided by the AI image generation service Midjourney

According to Reverie, the model has handled 3 million API calls in the past year, showing excellent accuracy and speed. In independent testing with Deepgram, Reverie's model achieved about 4.2% higher accuracy and a 1.5 times faster response time. This makes it a powerful speech recognition system for Indian users.

The advantage of this model lies in its ability to understand multiple languages and cultural backgrounds. Whether it's the number "twenty-three" said in English or "तेईस" said in Hindi, the model can accurately recognize it. Additionally, it can recognize names from all over India, taking into account differences in spelling and pronunciation, which are often challenging for global models.

Aside from Hinglish, Reverie has also launched a series of STT models for other Indian languages, including Tamil, Telugu, Bengali, Marathi, Gujarati, Kannada, Malayalam, Assamese, Oriya, and Punjabi. Each model is independently trained for specific dialects and accents of the language, aiming to more authentically reflect the way people use their local languages.

Pranjal Nayak, Reverie's R&D head, said: "Our research and development has always focused on unique linguistic challenges in India, and this Hinglish model is one of the results. It can understand the habits of Indians when using numbers, and how they mix English and Hindi in the same sentence." This makes AI agents' performance more natural and human-like.

This model has already been applied in multiple industries. For example, a large financial services company used Reverie's STT engine to process over 15,000 multilingual debt collection calls, successfully achieving high accuracy in recognizing numbers and payments.

Currently, this model is available on Reverie's API platform, and enterprises can use it through cloud or on-premise deployment. It supports language packages for specific fields, disambiguation of numbers and names, and hot word enhancement features, all of which can be configured through the same API.

Key Points:

🌟 This new model outperforms Deepgram in accuracy and response speed, meeting the needs of the Indian market.  

💬 Reverie's model can understand various mixed languages like Hinglish, and has a deep understanding of cultural context.  

📈 Multiple industries have started using this technology, significantly improving the accuracy and efficiency of speech recognition.