Lot of variation in generated voice through sockets | ElevenLabs | Page 1

jovial nimbus Apr 9, 2024, 6:14 PM

#

"voice_settings": {
"stability": 0.75,
"similarity_boost": 0.8
},

Using Sarah voice
model: eleven_turbo_v2

if you listen the attached recording, you'd notice how sweet the voice towards the end (last 20 seconds)

odd apex Apr 10, 2024, 6:41 AM

#

The length of your audio is too long, these variations happen the longer the audio being created is. I suggest you break your content into smaller chunks. Or use Projects as it is more stable in these cases..

jovial nimbus Apr 10, 2024, 6:48 AM

#

thanks @odd apex I am using websockets during a phone call to generate so looks like I d have to reconnect after every 30 seconds or so. should that solve the problem?

odd apex Apr 10, 2024, 6:49 AM

#

I'm not sure about that, seems like a hard thinkg to solve if you need to use Websockets..

jovial nimbus Apr 10, 2024, 6:52 AM

#

What about with stream?

#

POST request would be slow i guess

odd apex Apr 10, 2024, 6:54 AM

#

most likely

#

unfortunately apart from projects and even there, once you get over around 2000 characters in one api call you can start to see these variations..

jovial nimbus Apr 10, 2024, 6:56 AM

#

hmm in that case reconnect on websocket, should perhaps solve the issue?

#

the voice is getting extra sweet as we go long 😄

#Lot of variation in generated voice through sockets