GTM Sauce

Parloa · B2B · AI phone agents for customer service

Fine-tune speech models per country and customer to fix accuracy in small languages

Unknownlocalizationfeature-development

What they did

Built a 'language factory' product: an infrastructure that brings people onto the platform to label speech data for a specific enterprise, to fine-tune speech-to-text per country and customer (phone audio is worse than WebRTC). Also fine-tunes open-weight LLMs on the harness for specific use cases (matching or beating frontier models at lower latency and cost) and trains custom text-to-speech voices so a bank or insurer gets a voice no one else uses.
2:11
“We actually built an own language factory. It's an it's an own product we have where we fine-tune models based on a country and and a specific customer to increase the accuracy, especially on the phone where the quality of the voice is still is still subpar to like like WebRTC audio quality.”
Malte Kosub

Related topics

Get tactics like this every Monday

New tactics from the week's founder interviews, each linked to where it was said.

More from this episode