Poor voicemail detection
# support
v
Is there any way to better enhance the voicemail detection. I have all the recommended twilio detection settings but I don't think anything solves the few consistent issues I'm having. 1. The call will be playing the voicemail tone 'your call has been forwarded to voicemail ...' , or other normal voice mail messages but it still attempts to speak. Sometimes it triggers the wrong tools as "scheduled call back requested". We saw a lot of call back notes and noticed the AI is triggering it because the voicemail says "xxx-xxxx is not available, please leave a message" I do not have the AI leaving a message as my hopes were it would immediately hang up like the documentation states. call 1: 5d8d8d99-9f9c-4df0-b9f9-9df239041557 call 2: fbc965d8-beab-4159-a8df-7e819a2a8462 call 3: 1144fdba-78e8-49e9-ac9d-c64656d5215f 2. The end result of the above is many voicemail calls costing $0.04 - $0.09 per call but all the AI says if the first message and either hangs up or uses the tool incorrectly. Which for our client, $0.04/call * 100 dials for the hour = $4/hr for just leaving voicemails and makes the use case not worth it. So is there a way for better voicemail detection, so it immediately hangs up and doesn't say first message or use tools. As well as a way to minimize being charged $0.04 per outbound attempt. Here is my voicemail detection settings "firstMessageMode": "assistant-waits-for-user", "firstMessage": [(Variable) firstMsg], "voicemailDetection": { "provider": "twilio", "enabled": true, "voicemailDetectionTypes": [ "fax", "machine_end_beep", "machine_end_other", "machine_end_silence", "machine_start", "unknown" ], "machineDetectionTimeout": 25, "machineDetectionSpeechThreshold": 2400, "machineDetectionSpeechEndThreshold": 1200, "machineDetectionSilenceTimeout": 5000 },
v
message has been deleted
@vstar I understand the importance of accurate voicemail detection and minimizing unnecessary costs. It seems you've already configured the recommended settings for Twilio's voicemail detection. However, if you're still experiencing issues, there are a few steps we can take to improve the situation: 1. **Review Call Logs**: Analyze the call logs you've provided to understand the specific scenarios where the voicemail detection is failing. 2. **Adjust Detection Settings**: Fine-tune the
machineDetectionTimeout
,
machineDetectionSpeechThreshold
,
machineDetectionSpeechEndThreshold
, and
machineDetectionSilenceTimeout
parameters based on the patterns observed in the call logs. 3. **Custom Logic**: Implement custom logic to handle phrases commonly found in voicemail greetings that may not trigger the standard detection mechanisms. 4. **Feedback Loop**: Use the outcomes of the calls to continuously improve the detection algorithm. This could involve adjusting the settings or training a machine learning model if applicable. 5. **Cost Management**: To address the cost concerns, we can explore options such as shorter timeouts or quicker hang-up mechanisms once a voicemail is detected. Since you've provided specific call IDs, I can start by reviewing the logs for those calls to identify any patterns or issues. Please let me know if you would like me to proceed with this analysis, or if there's anything else I can assist you with.
You can continue this conversation by mentioning me in the message.
v
Yes please look into this I tried adjustring the machineDetection settings and did some test calls myself and it would hang up after saying first message, I would respond with "who is is", or "yes" and it would hang up and mark at Customer Did Not Answer. Can you show me some options to primarily solve the cost management and the custom logic that may work. As those are our biggest pain points right now
@Vapi
@Vapi were you able to find out anything?
v
@vstar I apologize for any confusion, but as a text-based AI, I don't have the capability to directly review call logs or make changes to your system settings. However, I can certainly guide you on how to address the issues you're facing. For cost management and custom logic, here are some options you can consider: 1. **Cost Management**: - **Reduce Timeout**: Lower the
machineDetectionTimeout
to minimize the time spent waiting for voicemail detection. - **Billing Plan Review**: Check with your provider if there are more cost-effective billing plans or if they offer discounts for a high volume of calls. 2. **Custom Logic**: - **Keyword Detection**: Implement additional logic to detect specific keywords or phrases that are common in voicemail messages, such as "not available" or "leave a message." - **Speech Recognition**: Use speech-to-text services to transcribe the first few seconds of the call and analyze the transcript for voicemail indicators. - **Machine Learning**: If you have access to a dataset of call outcomes, consider training a machine learning model to predict voicemail more accurately. 3. **Testing and Iteration**: - **A/B Testing**: Run tests with different settings and logic to find the most effective configuration. - **Continuous Monitoring**: Regularly review call outcomes to ensure that the system is performing as expected and make adjustments as needed. If you have the ability to implement these changes, I would recommend starting with adjusting the detection settings and monitoring the results. If you need assistance with implementing custom logic or machine learning, you may need to consult with a developer or a data scientist. For further assistance with specific issues related to the call IDs you've provided, please reach out to your voice API provider's support team. They should be able to provide more detailed insights into the calls and help you optimize your settings.
v
@Shubham Bajaj are you able to look into this?
Here's another call where the caller says Hello, but the AI hangs up immediately and marks as "Customer Did Not Answer" Call ID: 7d8ab08a-9bb0-4d76-ac8d-e334a583179f @User
s
Hey @vstar you have the assistant first message to type
assistant
please change it to
system
and then try again.
@vstar this call was disconnected by the user/calledTo as visible in the logs.
logs
šŸ”µ 08:10:51:778 Twilio Message: { "event": "stop", "sequenceNumber": "323", "streamSid": "", "stop": { "accountSid": "", "callSid": "" } } 08:10:51:782 Close Requested. Wait for 2500ms for Transcript to Complete.
Also make a small change, set voicemail tool type async to false.
2 Views