I use talkscriber as transcriber Elevenlabs voice (Jaffer) is not working. It does not reply back. What should I do ? Which voices should we try so that it works ? It does not work for us ? https://cdn.discordapp.com/attachments/1326211985758748703/1326211986052092015/Screenshot_2025-01-07_203140.png?ex=677e9aa9&is=677d4929&hm=f5fdabad0c0af1ee848fe328bccb052381fe10754ce0f914e0bbbda61e3decf7&

message has been deleted

it says You are using ar language for your transcriber. Make sure you choose a voice compatible with ar.

Got it. Tagging @User for additional help.

You are using ar language for your transcriber. Make sure you choose a voice compatible with ar.

@User What are the best practices for building a smart assistant in Arabic for outbound calls and customer interactions? I am using Talkscriber as a provider, the Whisper model for speech recognition, and 11Labs Rachel for voice synthesis. I have also implemented Eleven Multilingual V2, but the interaction between the user and the assistant is still not smooth. Can you provide recommendations to improve this?

Hey @Osman To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help me figure out what went wrong!

Hi @Shubham Bajaj , Thanks for reaching out! Here’s the information you requested to help track down the issue: I’ve noticed some specific issues with the assistant's interactions in Arabic: The assistant often delays responses, making conversations feel slow and unresponsive. It tends to stop speaking mid-conversation and resumes speaking in an inconsistent manner. The overall quality of communication in Arabic doesn’t match the fluency and professionalism exhibited in English. Here are the assistant’s technical details for reference: { "name": "Nova Assistant AR", "voice": { "model": "eleven_multilingual_v2", "voiceId": "21m00Tcm4TlvDq8ikWAM", "provider": "11labs", "stability": 0.5, "similarityBoost": 0.75, "fillerInjectionEnabled": false }, "model": { "model": "gpt-4o", "messages": [ { "role": "system", "content": "..." // Full assistant configuration details here } ], "provider": "openai", "temperature": 0.3, "emotionRecognitionEnabled": true }, "firstMessage": "السلام عليكم، أنا موظف من شركة نوفا، هل أتحدث مع الأستاذ {{4.

0

}}؟", "transcriber": { "model": "whisper", "language": "ar", "provider": "talkscriber" }, "clientMessages": [ "transcript", "hang", "function-call", "speech-update", "metadata", "transfer-update", "conversation-update" ], "serverMessages": [ "end-of-call-report" ], "backchannelingEnabled": true, "backgroundDenoisingEnabled": true, "startSpeakingPlan": { "smartEndpointingEnabled": true }, "stopSpeakingPlan": { "numWords": 3 } }

Because of no response from @Osman closing this ticket.

@Shubham Bajaj What are the best practices for building a smart assistant in Arabic for outbound calls and customer interactions? I am using Talkscriber as a provider, the Whisper model for speech recognition, and 11Labs Rachel for voice synthesis. I have also implemented Eleven Multilingual V2, but the interaction between the user and the assistant is still not smooth. Can you provide recommendations to improve this?

Hey @Osman To help track down this issue, could you share: - The call ID - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help me figure out what went wrong!

@Shubham Bajaj Call ID: 312f6293-d061-4ae8-977c-51071905e38e > The accent used is High Arabic (akin to Shakespearean English), which feels overly formal. > It’s in Egyptian Arabic, which isn’t suitable for audiences in Saudi Arabia or Dubai (this is critical). > Latency during the call was unusually high.

@Osman To meet your expectations, I believe for authentic Arabic you can either use a custom voice or a custom voice provider which supports Arabic as per your required accent. Regarding latency, I suggest setting the startSpeakingPlan waitSeconds to 0.8.

@Shubham Bajaj I need to implement this functionality specifically for Saudi Arabic. Which transcriber and voice would be best suited for this, and what settings should I use to achieve optimal results?

Preferences: Deepgram and 11labs. @Osman have you tried using them?

@Shubham Bajaj yes i tried with Deepgram and 11labs but the transcriber didnt work .Currently, Talkscriber with Azure voices is functional for me, though not ideal.

@Osman If that's the case then you have to bring your own custom-voice and custom-transcriber provider, just confirm with if they support streaming and then rest I can help you out with it.