My name is Daniel, and I’ve recently watched many of your videos — really great content.
After seeing your video about Retellia, I decided to try it out. I purchased a number and everything worked well until I upgraded to the Pro plan. Since then, I’ve noticed issues with the speech-to-text (STT) system, specifically when the conversation is in Spanish.
It seems to struggle with proper names in Spanish and short confirmations. For example:
When I say “Ainhoa”, it gets transcribed as “Ahí no”.
When I say “Ignacio Pila”, it is interpreted as “Ahí nació Pilar”.
It also has trouble correctly recognizing simple responses like “sí” or “no”.
Example conversation:
Agent (0:12): Perfecto. ¿Cómo te llamas?
User (0:15): Ahí no.
I would like to know if this is a common issue when using Spanish, and whether there are any settings, configurations, or best practices to improve recognition accuracy for names and short answers.
Some of the key settings you can check are as follows.
Boosted Keywords: Add the proper names (e.g., “Ainhoa”, “Ignacio”, “Pila”) as custom keywords. This enhances the STT model’s vocabulary — you can add up to 100 keywords.
Denoising Mode: Short responses like “sí” or “no” are often filtered out by denoising. Set denoising mode to “No Denoising” to preserve short, low-energy utterances.
Transcription Mode: Switch to “Optimize for Accuracy” — it waits longer for more context, which helps with entities like names (~200ms extra latency).
Language Setting: Ensure your agent’s language is explicitly set to Spanish so the STT model expects Spanish input.
Please check these settings first. If the issue still persists, kindly share a call ID where you experienced the problem so we can investigate further.
hey , this is one example call_f2eb4e603128d739c00fc5475d0 , where the name is aihoa pirqueras , and it got Ai for the name , and in here i said multiple times no and it got the lastone thanks.