i.
First touch
“Hi! Thanks so much for calling. I see you're having trouble with your order. Let me pull that up real quick and we'll figure it out together.”
0:00
· CodaThe Coda demo
Be the voice
customers actually
want to talk to.
Natural-sounding, accurately-pronouncing TTS built for high-stakes enterprise conversations, where every word counts. Listen to Coda across a few real shapes, then talk to a live agent yourself.
Talk to it · right now
live agentwawona · coda
Rime sales engineer
Ask anything about Rime — voices, latency, on-prem, custom builds.
live · rime · coda · livekit cloud
mic access required · gpt-4.1-mini · deepgram nova-3 · rime coda · livekit cloud
A live Rime sales engineer in voice form. Ask about voices, latency, on-prem, custom builds — anything.
Results · publicly reported
The right voice drives
better business results.
42%+
Fortune 500 client
lift in customer engagement after switching to Rime voices.
10%+
ConverseNow
improvement in restaurant order accuracy and resolution.
15%+
Domino's
gain on key conversational quality benchmarks vs. prior TTS.
01 · Use case · contact center
The anatomy of a support call.
The premise
A single voice carrying the whole arc of a real customer-support call — greeting, diagnosis, resolution, the unexpected wrinkle, and the close. Five clips, one continuous shape. The test isn't whether one line sounds good. It's whether the same voice still feels like a person five minutes in.
Samples · wawona · coda
5 clipsii.
Reading the problem
“Okay, I'm looking at your order right now. Looks like the shipment got delayed at the warehouse. Honestly, that shouldn't have happened. I'm really sorry about that.”
0:00
iii.
Offering a way out
“Here's what I can do for you. I can either get a new order out today on overnight shipping, no extra charge, or I can process a full refund right now. Whichever works better for you.”
0:00
iv.
Catching the wrinkle
“Oh, hold on, I'm seeing one more thing here. Looks like the tracking had an older address on it. Can you confirm your current shipping address so I get this right?”
0:00
v.
Wrapping it up
“Got it. So we'll have that out today, you should see it tomorrow morning. Anything else I can help with? Thanks so much for your patience, really appreciate it.”
0:00
02 · Use case · companion / consumer
Same script, four voices.
The premise
Hearing the same lines across multiple voices is the most honest A/B you can run. The script stays still; what changes is the person behind the words. Pick the warmth that fits your product, then we'll build around it.
Samples · vespera · arcade · celeste · marlu
5 clipsi.
Vespera — the first hello
“Hi, I'm so glad you called. So, this is the part where we just get to know each other a little. There's no script, no right answer. I'm just gonna ask you a few things, and we'll see where it goes. Sound good?”
0:00
ii.
Arcade — the first hello
“Same script — different read. Bright, curious, present.”
0:00
iii.
Celeste — asking something real
“Okay, so most apps want to know your height and where you went to school. I'm more interested in something else. What's been on your mind this week?”
0:00
iv.
Marlu — asking something real
“Same script — different read. Easy, intimate, conversational.”
0:00
v.
Vespera — holding what they said
“Hm. The way you said that. It sounded like it actually matters to you. You know, I think the people who pay attention to things like that are usually the ones who care about the right stuff.”
0:00
03 · Use case · one voice, many desks
Three reads of a single voice.
The premise
The same voice handles a warm welcome, a tight IVR prompt, and a loose conversational agent — without sounding like three different recordings. Useful when one brand voice needs to cover the whole contact center.
Samples · marlu · coda
3 clipsi.
Warm — the welcome
“The line between professional and friendly — the default front-desk read.”
0:00
ii.
Brief — the IVR
“Tight, low-affect delivery. The kind of read a menu prompt actually wants.”
0:00
iii.
Conversational — the agent
“Loose, contraction-heavy, like a real rep on the line.”
0:00
Deploy your way
On-prem or in the cloud.
Same model, same voices, same plugin — wherever you need it to run. Hosted in two regions today, or shipped as a container into your own infrastructure.
Hosted cloud
us-west-2 (default) + us-east. Same-region calls land in 1–10ms.
On-prem container
Drop-in deploy. ~140ms TTFB on a single H100 or RTX 6000.
Two integrations
First-party LiveKit Agents plugin + a direct HTTP/WS API.
About · Rime
Built for calls you
can’t afford to get wrong.
What Rime is
- Coda — current flagship model. Discrete-token architecture, built for conversational latency and expressive prosody. Everything on this page is Coda.
- Mist v3 — the production workhorse. Phoneme-first, deterministic pronunciation, ~140ms TTFB on an H100 or RTX 6000.
- ~55 voices, mostly US English plus en-GB, en-AU, en-SG, en-DO. Custom brand voices in the Coda family ship in roughly 3–4 weeks end to end.
- Two regions, us-west-2 (default) and us-east. Same-region calls land in 1–10ms.
- Two deployment modes — hosted cloud and on-prem container. On-prem TTFB is the same ~140ms on a single H100.
- An in-house linguistics team shapes prosody, text normalization, and pronunciation rules per domain. We are the only TTS company built this way.
- In production at scale with teams including DoorDash, who run Rime in their voice-agent stack today.
Drop it in
Rime ships a first-party LiveKit Agents plugin. Two lines and the voice changes.
# install
pip install "livekit-agents[rime,silero,turn-detector]~=1.3"
# env
export RIME_API_KEY=...
# agent.py
from livekit.agents import AgentSession
from livekit.plugins import rime
session = AgentSession(
# ...vad, stt, llm
tts=rime.TTS(
api_key=os.environ["RIME_API_KEY"],
model="coda",
speaker="wawona",
lang="eng",
),
)livekit-agents 1.3 with the rime extra · python 3.10–3.13