IndiaSpeaks Research Labs Private Limited
About Company
We are a team of passionate researchers and engineers set out to make businesses accessible to common man through VoiceTech. We develop language technology tools for Indian languages enabling businesses reach out to their customers in local languages. Our products use proprietary voice recognition, text-to-speech and natural language processing tools. Customer acquisition and retention is a known and challenging task for any business especially when expanding to the Tier 2+ cities of India where language acts as a main barrier in building trust and on-boarding the next 500 million users from these regions who are entering Internet/Mobile space for the first time. Even though advanced technologies are growing and businesses making products and services easily accessible, these benefits remain unknown to large chunk of Indian population due to unavailability of content and service in their local languages.Almost 75% on India?s population can read, write and speak in their native language and around 80% of Indians above 15 years owns an account at a bank or an other financial institution. To serve this substantial customer through a digital channel, banks certainly need to innovate and serve them in the regional language of the customer. Furthermore, for e-commerce to penetrate away from cosmopolitan cities, they must use these technologies to understand customer needs, address their queries, structure and analyze the data generated by the customers, etc. Many feature-driven markets such as electronics and cars where even a small feature may sway customer decision, are yearning for these features to gain (or maintain) the competitive edge. For example, BigBasket uses Google?s speech recognition engine with minimal functionality for product search. MG Hector uses few voice commands in Hindi. Multiple entertainment platforms use content descriptions and listing in Indian languages. With immense scope for both technological development and product maturity, market readiness plays crucial role in making this technology ubiquitous in public life across our country.
Deployment Details (Projects/ Pilots/ Field Trials)
We have worked on Pilot project demo for Delhi PoliceWe have deployed demo in BBNL/FoFi IOT Labs for Channel navigation through voice supportWe are also working with HedgeLoop on a chatbot projectPatents and IP disclosures
Indian patent: Methods for creation of subword dictionary Application number: 202141025895 Summary of the patent: The present invention discloses a method for creating and modelling a subword dictiona
List of Key Offerings
Products and Solutions
We offer customized language models
vocabulary
chatbots and dialogue flow tailored as per the business needs and applications. Our services are available in English
Hindi
Tamil
Telugu
Kannada and Malayalam. Soon to serve other Indian languages as well. Our voice services can either be deployed offline or on-cloud. Light-weight models for Android
iOS platforms are also available. We also provide valuable insights from consumer interactions through our speech analytics engine. With over a decade-long experience in the field of speech processing
we founded IndiaSpeaks in 2021 and have covered a significant ground in fulfilling our goal to enrich lives of millions of end-consumers across India. We have served multiple businesses across verticals like Entertainment
EdTech
FinTech and call centres. We have a robust semi-automated data collection pipeline that handles text selection
voice quality check and curation.Products/Services overviewAll the tools that we provide will be highly efficient state-of-art ML and AI models that will be trained with the speech and text data collected through our semi-automated data collection pipeline. These ML models will be tested for robustness against all adversarial conditions before deployment. The details about our various products are given below.Voice recognition systemsThe main purpose of a voice recognition system is to convert a given voice signal to its corresponding orthographic text which will then be used by the subsequent NLP systems. A good voice recognition system should be robust to channel
speaker
age
gender
environment
noise and recording device variations present in the voice. We will periodically improve the accuracy of our voice recognition systems by mainly focusing on the following factors:? Increasing the amount of training data
typically in the order of thousands of hours.? Improving the quality of the ground truth data (a.k.a the audio transcriptions).? Diversifying the nature of the training data with respect to the speaker?s age
gender
accent and dialect regions.? Improving the complexity
architecture and training duration of the ML models.? Handling the morphological complexity and effective vocabulary building for thelanguages.Text-to-speech systemsText-to-speech (TTS) is an alternative assistive technology that allows end users to consume contents that a business provides online. It enables the content to be audible so that a larger pool of users have personalized and humane experience. This technology will primarily benefit the semi-literate Indian users who can now seamlessly access the application without any difficulties. Further
TTS can be thought of as the voice of any business (e.g. Apple?s Siri and Amazon?s Alexa). The voice of a business? TTS could psychologically bring instant connection between the user and the business
and so it is important for any business to have a real-time
high-quality
appealing
human-like synthetic voice that interacts with the end users. Our startup offers such multi-speaker
high-quality TTS systems that could potentially become the voice of our customers? businesses. Natural language processing systems Natural language processing (NLP) systems enable machines the ability to read
analyze
understand and derive meaning from spoken or written form of human languages. These systems process the input text from voice recognition system and identifies the intent
context
name-entities and sentiments to automatically trigger actions. Such systems can be built efficiently by training on large amount of annotated text data. Our product offerings include NLP systems that can be used to build chatbot
appointment booking
question-answering
sales lead generation
bill payments
request processing
etc. We will develop the NLP modules taking into account the complex morphological structure of the Indian languages.Chatbot systemsChatbot represents the natural evolution of a question-answering system leveraging various NLP modules. The complexity of a chatbot varies from a simple plug-and-play question-answering system to dynamic state machine knowledge graph-based systems. According to the nature of the clients? businesses
these chatbots will be developed to specifically cater to their needs.The output from the chatbot system is passed through the content resource management (CRM) systems which executes a series of actions and its responses are played back to the end-user via TTS. In this new era of AI
chatbots can be considered as an effective replacement to highly human-intensive call centre operations where tasks can be automatically performed with predefined templates and workflows. Our business solution based on chatbot includes question-answering
ticket raising based on complaints
