Generating insights from call center data via speech-to-text analytics (aka voice-to-text) is a hot topic and increasingly important, but it’s still not that easy. Technology allows us to capture everything, and customers know it. That means they expect response and adjustment – action based on the feedback they provide, no matter where they provide it. This poses not just a challenge, but also an opportunity.
Customer-centric businesses are under constant pressure to understand, satisfy and delight their customers, and the call center is one of the most valuable – and also most difficult – sources of input to process. Gleaning insights from call center data has historically been expensive and difficult to automate, because of three primary components:
Audio files are inherently complex, with at least two people (agent and caller/respondent) talking. Optimally, the data would be captured via dual-channel mode in a diarized file, in other words, record the comments for processing in the order in which they occur and with each party distinctly captured. This is important to enable the text analytics solution to distinguish the customer from the company and question from answer.
Quality of the audio files is tricky to preserve. This kind of data is like every other kind – garbage in, garbage out – but more so. Background noise and bad connections can make it difficult for machines to separate the actual words from the other sounds. Keen focus on recording methods and quality is essential to being able to process the data later.
Files often get compressed to save space. File compression may seem like a good idea, especially with large volumes of audio data, but this makes the data impossible for software to process. To allow for speech-to-text / voice-to-text analysis, preserve the uncompressed audio files (.wav) or pulse code modulation (.pcm) files in their original format.
Once you capture, collect and store the data in the right format with the suitable quality, there are tremendous opportunities to import and process the data automatically via speech-to-text analytics, enabling:
Speech-to-text analysis opens up a major opportunity to conduct Sentiment Analysis on unstructured data from call centers, not just for what people are talking about, but also for how they feel about it, and then further to aggregate the data to find patterns that provide insight as to why they feel that way or how different sub segments respond differently.
When your customers are talking to customer service, they are sharing what they need, want, like and don’t like in their own words on their own time, creating a valuable, unfiltered supplement to your traditional CX surveys. Automated speech-to-text analysis increases your ability to capitalize on customer feedback and actionable insights to drive continuous improvement, innovation, customer satisfaction and loyalty.
With the data connected, analysis automated and insights flowing, it is possible to push the automation upstream to enable real-time insight delivery closer to – or even during – the call. This would involve building language models based on a sample for more sophisticated telephony integration via dynamic, closed-loop call management.
Contact us to discuss your voice-to-text or speech-to-text analytics inquiries and needs.
You can easily pull your API survey data and even voice-to-text data into the Ascribe Intelligence suite for faster, more comprehensive access to actionable insights.Find out how you can get access