Difference between total character durations and buffer duration

[lbaliunas]

Hi,

I'm using ElevenLabs websockets. I'm using ulaw_8000 output format, and get the alignment times for each chunk. I notice that there is a difference between audio buffer duration, calculated by dividing the number of bytes by 8000 (as per ulaw_8000 encoding), and the sum of the charDurationsMs provided in the alignment.

Could I get clarification on why is that so? Shouldn't they be the same?

Thanks!

[louisjoecodes]

Hey @lbaliunas apologies for the delay here.

Is the delta between the alignment timings significant? Does it impact the accuracy of the alignments at word-level for example?

Curious to know if/how you solved this.

[lbaliunas]

Hi @louisjoecodes , it depends, but it varies to both sides (sometimes less sometimes more than the actual duration of the buffer). The issue is that I'm trying to track "how much was said" when the audio is stopped, and it's difficult to use alignment timestamps when they don't add up to the actual buffer duration.

Any ideas where these discrepancies appear? Maybe in the pauses/whitespaces?

[PaulAsjes]

Would you be able to provide some examples of the discrepancy? Would be helpful to see logs.