Recommendation - OpenAI Whisper Vmail Transcription is Excellent

[Lee Cramman]

Not a question as such, just to say that we used to have problems with Google's voicemail transcription to the point it was pretty much useless to us (other than laughing at the results).

We have to handle lots of messages from patients about medications with long, complex names as well as lots of quite niche medical terminology (and not to mention that the local dialect is strong, and makes it hard for voice recognition) and Whisper handles it like an absolute champ. Words like levothyroxine, hydroxocobalamin, clopidogrel, infliximab - most of it is correct right out of the gate, and the odd error is handled gracefully whereby you can tell what the person was trying to say without having to listen back to the message.

Pricing is very reasonable too - I would recommend it to your customers who have struggled with transcription quality in the past.

 

[VoIPTools]

I have to concur 100% with @Lee Cramman comments.

As the developer of 3CX Voicemail Manager, we were the first to include voicemail transcriptions for 3CX in our products (about 6 years before 3CX added the feature). Over 15+ years I have personally tried countless voice-to-text engines for transcribing 3CX voicemails. Every engine we have tried, including some ridiculously expensive solutions costing tens of thousands of dollars a month, have all struggled with the same problem -- the modest audio quality used by 3CX for voicemails and recordings.

To a human, the original 3CX audio quality is perfectly acceptable, but every engine we tried all failed, without exception, to provide a accurate transcription... until OpenAI.

The audio quality has always been the challenge. In my tests, for example, I used the same microphone to record the exact same message using 3CX, and the simple "voice recorder" built into Windows. I then processed both recordings. Almost every engine we tried could transcribe the voice recorder version of the message with very acceptable accuracy -- about 90+% on average. But the 3CX recording was typically only about 75% accurate -- accurate enough to get a general intent of the phone call, but often times you sill needed to listen to the recording to fully grasp the purpose of the call. This pattern held true regardless of the engine. Even worse, heaven-forbid if the caller had a difficult accent, or a lot of background noise. In that environment the accuracy would fall to completely unusable.

But that has all changed with OpenAI. OpenAI has literally no trouble transcribing 3CX voicemails and recordings. The accuracy is generally perfect or nearly perfect regardless of the background or accent (within reason). I cannot understate what a change this technology represents. If you have not tried the new transcription feature in 3CX V20, it is time you try it. It is absolutely amazing. This is a rare example where the technology actually matches the hype. And there is a lot more coming in future versions of 3CX and in VoIPTools. Wait until you see our new 3CX Recording Manager tool. You will be blown away.

 

[smillstid]

I have to concur 100% with @Lee Cramman comments.

As the developer of 3CX Voicemail Manager, we were the first to include voicemail transcriptions for 3CX in our products (about 6 years before 3CX added the feature). Over 15+ years I have personally tried countless voice-to-text engines for transcribing 3CX voicemails. Every engine we have tried, including some ridiculously expensive solutions costing tens of thousands of dollars a month, have all struggled with the same problem -- the modest audio quality used by 3CX for voicemails and recordings.

To a human, the original 3CX audio quality is perfectly acceptable, but every engine we tried all failed, without exception, to provide a accurate transcription... until OpenAI.

The audio quality has always been the challenge. In my tests, for example, I used the same microphone to record the exact same message using 3CX, and the simple "voice recorder" built into Windows. I then processed both recordings. Almost every engine we tried could transcribe the voice recorder version of the message with very acceptable accuracy -- about 90+% on average. But the 3CX recording was typically only about 75% accurate -- accurate enough to get a general intent of the phone call, but often times you sill needed to listen to the recording to fully grasp the purpose of the call. This pattern held true regardless of the engine. Even worse, heaven-forbid if the caller had a difficult accent, or a lot of background noise. In that environment the accuracy would fall to completely unusable.

But that has all changed with OpenAI. OpenAI has literally no trouble transcribing 3CX voicemails and recordings. The accuracy is generally perfect or nearly perfect regardless of the background or accent (within reason). I cannot understate what a change this technology represents. If you have not tried the new transcription feature in 3CX V20, it is time you try it. It is absolutely amazing. This is a rare example where the technology actually matches the hype. And there is a lot more coming in future versions of 3CX and in VoIPTools. Wait until you see our new 3CX Recording Manager tool. You will be blown away.

If you have recordings enable for this transcription, are they simply emailed directly to the user?

 

[VoIPTools]

In our tool, you will get the transcription, summary, key points, action items, appointments, and whatever other OpenAI prompt you specify in our software. Give me your wish-list and I will make it happen. We are shooting for release on Friday.