#"speed" and "emotion" experimental controls seem to make no difference to output (on websocket)

1 messages · Page 1 of 1 (latest)

serene patio
#

Hey Cartesia,

I'm working on an initial implementation of Cartesia, and have run into an issue. It seems like the "speed" and "emotion" experimental controls have no impact on the synthesised audio. I'm using a websocket connection and the TS library:

const response = await websocket.send({
modelId: process.env.CARTESIA_MODEL_ID!,
voice: {
mode: "id",
id: process.env.CARTESIA_VOICE_ID!,
experimentalControls: {
speed: "slowest",
emotion: []
}
},
transcript: text,
addTimestamps: true,
});

I've seen a couple of other posts that suggest removing the "emotion" parameter. I've confirmed that the following is sent at a low level:

{
"model_id": "sonic-2",
"voice": {
"mode": "id",
"id": "a01c369f-6d2d-4185-bc20-b32c225eab70",
"__experimental_controls": {
"speed": "slowest",
"emotion": []
}
},
"transcript": "<our text>",
"add_timestamps": true,
"context_id": "easy-doors-guess",
"output_format": {
"container": "raw",
"encoding": "pcm_s16le",
"sample_rate": 8000
}
}

I've tried with a couple of voices with the same result. Is this control expected to be working currently?

Thanks in advance,
Joe

worn stream
#

@serene patio you need to define a slightly older version of sonic-2 as the latest version does not allow experimental controls

#

try swapping from sonic-2 to sonic-2-2025-03-07

small yew
#

Correct

#

We are deprecating experimental controls in the model

sick frost
small yew
#

Yes, we would suggest choosing a different voice as a fix

tranquil scaffold
#

So, there is now way to make slow any voice. even the future. right?

Then how can I make a slow voice?

serene patio
#

Thanks @small yew, really appreciate the response. Given we're UK-based, we don't have many voices to choose from, and this one was basically the only one that matched our brand, but it is very fast for our user demographic (generally 70+ years old).

#

I'll try it out with an older model

cobalt pulsar
#

This is a bummer. I'm using the speed controls for a language learning use case so learners can hear things more distinctly on repeat if they need to.

hot dove
#

Hey! We're actually bringing back a version of speed controls for slow, medium, and fast in the next few days! Will let you all know when it's out 🙂

tranquil scaffold
#

Wow! Thank you!

tranquil scaffold
#

the speed option is returned. (but different 🥹)

image0.jpg
hot dove
#

Hey! Yes this is out! It's less pronounced than before, but will still make a difference (and be stable at all speeds). The pacing overall on voices have also been improved and are more like the original now.