Hi! Is there a way to correct pronunciation of TLDs? For example, we collect inbound emails, and then when it's

.ca

the AI responds with

.k

(as how it's pronounced as the letter k)

You can continue this conversation by mentioning me in the message.

Awesome, yeah that'd be great. Right now I'm just using

deepgram

in the UI, but I don't see any config options for this. We are invoking the assistant via API

@HashTrade Great, thanks for clarifying that you're using ElevenLabs for voice synthesis. To correct pronunciation with ElevenLabs, you can use their API to specify pronunciation rules or provide phonetic spelling for certain words or phrases. Since you're invoking the assistant via API, you would typically include pronunciation instructions in your API request. ElevenLabs may offer a way to define how specific words or acronyms should be pronounced, possibly through SSML or a similar markup language. Unfortunately, I don't have access to the specific documentation or methods for ElevenLabs, but I would recommend checking their API documentation or reaching out to their support team for guidance on how to implement custom pronunciation for TLDs like

.ca

. If you need further assistance with this or have any other questions, please let me know!

Right, so my understanding, or my take would be that this isn't currently available? @Vapi As we don't really invoke the elevenlabs API directly ourselves in your workflow

@HashTrade I apologize for any confusion. If you're not directly invoking the ElevenLabs API and are using an intermediary service or workflow that integrates with ElevenLabs, then the ability to correct pronunciation would depend on whether that intermediary allows for customization of the text-to-speech output. If the intermediary service you're using with ElevenLabs does not provide an option to customize pronunciation, then it might not be possible to correct the pronunciation of TLDs like

.ca

within your current setup. However, if you have the ability to modify the text input before it's sent to the voice synthesis engine, you could try spelling out the TLDs phonetically (e.g., "dot cee ay" for

.ca

) in the text that's being converted to speech. If this is not an option and you require further assistance, I would recommend reaching out to the support team of the intermediary service or ElevenLabs for more specific guidance on how to address this issue within your workflow. Please let me know if there's anything else I can assist you with!

Can you raise this concern with their team?

@HashTrade Please ask support for more info on this.

@HashTrade you have to work on your prompt, such as asking user to spell out email character by character and then confirm it back with the user.

example

- Could you please share your email address, one letter at a time, so I can make sure I get it right? - - Once the user provides the email, confirm: Let me repeat that back to double-check everything is correct.