Statistical Speech Recognition for Virtual Assistants
Advanced data analytics and natural language processing techniques are becoming a key focal point for investment banks and brokerage firms.
Average percentage of the day a typical sales and trading professional spent on administrative tasks
Average percentage of the day a typical sales and trading professional spent on actions related to their official job function
Funding raised to further develop the client's virtual assistant and market intelligence platform
Our client is an early stage growth company backed by two of the world's largest investment banks and an innovative venture capital firm. They recently raised funding to develop an end-to-end software solution to help sales and trading professionals within the investment banking and brokerage sectors to reduce time spent on administrative tasks and give them more time to focus on strategic initiatives that are important to their success.
According to data provided by the client, the vast majority of sales and trading professionals spent roughly 54% of their time on administrative tasks, 29% analysing market information and 17% acting on strategic initiatives. We were asked to help the client improve the quality of their existing speech recognition systems using a variety of machine learning techniques. Our work formed part of a project to create a virtual assistant to make it easier for sales and trading professionals to reduce the amount of time spent on administrative tasks.
A specialist team of data scientists with recognised expertise in statistical signal processing, natural language processing and machine learning divided the analytics into three sprint sessions. The first session involved statistical signal processing for speech recognition systems with a strong focus applied to reducing background noise from other people on the trading floors and increasing the signal-to-noise ratio for the voice commands. The second session focused on the creation of natural language processing algorithms to translate the voice commands and audio data into written text. The third session focused on statistical quality control and machine learning, allowing sales and trading professionals to over-write any incorrect commands in the dictated text.
Our team of data scientists delivered a world class solution in collaboration with our client. Following delivery of our project the client signed a multi-year contract with the world's largest investment bank and become the first graduate from the bank’s ‘In-Residence’ fintech programme.
Note: Delivered in partnership with Pivigo Limited.