Choosing a suitable voice

If you’re building a voice-based agent, you'll be billed credits for two things:

• The number of minutes that the call lasts
• The text-to-speech costs used to generate the agent’s voice

Just like with LLMs, different voices cost different amounts of credits. Generally, more modern, human-sounding voices - such as those offered by 11labs and Rime - cost more credits than older, more robotic models such as those offered by Amazon and Google.

‍

You can change your agent’s voice in the settings section of the Voiceflow CMS.

‍

‍

Which voice should I choose?

While 11labs and Rime voices are significantly more expensive than legacy models, such as Amazon’s, we still recommend using them for customer facing production models. These human-sounding voices help customers feel comfortable trusting your agent, and create a better overall experience.

‍

However, there are some scenarios where you might want to switch to an Alexa, Google, or Microsoft voice. For example, if you’re prototyping and testing your agent a bunch from inside the editor, temporarily switching to one of these voices will significantly reduce your credit usage - just don’t forget to switch it back when publishing to production. Additionally, if you’re building an internal-only voice agent for your employees that’s only completing simple tasks and asking set questions, you probably don’t need the power of these more advanced voices.

‍