Hello, im currently working on a project and the assistant is set to ask name and document number. The problem is it understands and captures only part of the number. Exemple the person says the document number is 1234567. It only captures 1234. How can i correct that?

@nicholassantoss Hello! It sounds like you might need to adjust the settings of your assistant's transcriber to better capture longer sequences of numbers. Here are a few steps you can take to improve the accuracy: 1. **Check the Transcription Model**: Ensure that the transcriber is using a model optimized for the type of content you are transcribing. For example, some models are better at recognizing numbers or specific jargon. 2. **Adjust the Audio Quality**: Make sure that the audio input is clear and of high quality. Background noise or low-quality audio can affect the transcriber's performance. 3. **Custom Vocabulary**: If the transcriber supports it, add a custom vocabulary that includes the types of numbers or terms you expect to be captured. 4. **Post-Processing**: Implement a post-processing step where you can use regular expressions or other methods to validate and correct the transcribed numbers. 5. **Feedback Loop**: Use the incorrect transcriptions as feedback to improve the model if your system supports continuous learning. If you provide me with more details about the assistant or the transcriber you are using, I can give you more specific advice. If you're using a

DeepgramTranscriber

, you might want to look into its settings to see if there's an option to improve number recognition.

how do i do that on a live call?

@nicholassantoss To improve number capture during a live call, you can take the following steps: 1. **Real-time Feedback**: Implement a system where the assistant can ask for confirmation or repetition if it detects that the number capture might be incomplete. For example, "You mentioned the number 1234, is that the complete number?" 2. **Chunking**: Break down the number into smaller parts. For instance, ask the user to provide the number in chunks, like "Please say the first four digits followed by the next three digits." 3. **Improve Speech Recognition**: Use a speech recognition engine that is optimized for telephone audio and has robust number recognition capabilities. 4. **Signal Processing**: Apply noise reduction and echo cancellation to improve audio quality before it's processed by the speech recognition engine. 5. **User Instruction**: Guide users on how to speak their numbers, such as speaking slowly and clearly, which can help the system capture the full sequence. 6. **Adjust Timeout Settings**: Ensure that the speech recognition engine's timeout settings are not cutting off the user prematurely. If you're using a specific API or system for live calls, please provide more details so I can give you a more tailored solution. If there are settings or configurations within your Voice API that can be adjusted for better number recognition, those should be explored as well.

@nicholassantoss can you check the logs about what’s the transcriber output?

during live calls, the transcriber works lively?

yeah

I guess it works now

create a new support ticket for it with expected vs actual response with timestamps and call id.