Speech enhancement app for call centers
Problem?
Contact centers maintain dedicated quality assurance and coaching teams, but only a fraction of their calls can be reviewed. Feedback arrives late, training is difficult to personalize, and high attrition forces teams to repeat the same work for every new batch of agents.
Why on device?
A cloud API would add another network dependency to a live call. Running the models locally kept processing close to the audio path and made the product easier to deploy inside existing call center setups where they also have different levels of network firewall restrictions depending on the process policy.
What did I build?
I built a Tauri desktop application that packaged on-device voice models for live calls. The work joined model inference, audio routing, application releases, and software distribution into something teams could deploy without rebuilding their dialer.
What changed?
The application reached 20+ BPOs and call centers and increased call conversion by 58%. More importantly, it moved the models out of a demo and into the environment where agents were already working.
Hear the difference
Listen to the microphone input, then compare it with the enhanced output.
Original
Unprocessed microphone input
Enhanced
Speech-to-speech model output