How Talkpal cut TTS costs by 40% while improving retention and free-to-paid conversion
Talkpal is a conversational AI language learning platform with a mission to make high-quality language learning accessible to every learner worldwide. The platform teaches over 80 languages to more than 5 million users through real-time spoken practice, using AI to emulate real-life scenarios in which users learn by interacting with an AI teacher.
The language learning experience is inherently voice-intensive and with Talkpal’s rapidly growing user base, the company needed a text-to-speech (TTS) solution with broad multilingual support, low latency for natural conversation flow, voice variety to match diverse AI teacher personas, and competitive pricing.
Talkpal chose to use Inworld TTS across their platform, powering multilingual AI teachers, role-play characters, and various other learning modes for both its free and paid users.
Our mission is to make high-quality language learning accessible to every learner worldwide through cutting-edge AI and machine learning. Inworld’s reliable, cost-effective TTS technology directly supports this vision, and our partnership plays an important role in helping us scale that mission further.
Dimitri Dekanozishvili, Talkpal AI Co-Founder
Requirements of real-time language learning at scale
Building a leading conversational language learning product with voice at its core presents specific quality, technical, UX, and cost requirements that determine whether the product can scale.
Users practice speaking with AI teachers in real-time, where low-quality voices and small delays break the immersion that makes conversational learning effective. The AI voice experience must be natural and real-time so that learners stay engaged.
Conversational language learning also relies heavily on voice. Every conversation, pronunciation drill, and role-play requires TTS, meaning voice AI typically represents one of the largest infrastructure cost categories.
Talkpal also needed to support a wide range of languages and voice personas. Each AI teacher must have a distinct personality and requires a voice that matches. This required both multilingual capabilities and the ability to create custom voices through cloning.
Building for accessibility at scale with Inworld TTS
Talkpal integrated Inworld TTS across both paid and free user segments for various languages, including English, German, and French, and across different learning modes and languages.
Integration took less than a week and leverages:
- Inworld TTS-1 for high-quality, low-latency, multilingual voice AI
- Voice cloning for custom AI personas that match character designs
To validate the solution in a real-world environment, Talkpal ran a four-week A/B test and observed measurable improvements across key metrics:
- 40% reduction in TTS costs
- 7% increase in feature usage
- 4% lift in user retention
The improved voice quality for free users also drove an increase in free-to-paid conversion rate.
We chose Inworld because of its low latency, high-quality output, multilingual support and competitive pricing. In addition, with high rate limits and stable performance, Inworld has enabled us to scale confidently. Even during peak usage days, we’ve experienced no rate-limit issues or latency bottlenecks.
Dimitri Dekanozishvili, Talkpal AI Co-Founder
Scaling the mission
For Talkpal, voice AI is central to their mission of making high-quality language learning accessible to every learner worldwide. By reducing costs while improving the experience for both free and paid users, Inworld TTS helps Talkpal further that mission.
For teams building real-time conversational AI at scale,
talk to our team about your use-case.