Please rotate your device

To continue exploring this website, please rotate your device to a vertical orientation

Rime Mist v3

Rime/Rime Mist v3

Rime Mist v3 is an updated inference engine for the Mist text-to-speech model.

text to speech

Concurrency & Performance

Mist v3 is an ultra-low latency TTS model for real-time systems and AI agents, with TTFB low as 40 ms on GPUs like Ada Lovelace (L40S) and Blackwell (RTX PRO 6000), making it well-suited for steady-state production loads without complex auto-scaling. For agent-native companies where TTS is often the throughput bottleneck, this concurrency headroom is particularly significant.

On-Prem Deployment via Baseten

Existing Rime on-prem licensees can bring their Rime license to Baseten and get scalable inference on day one, without sub-processor onboarding or additional vendor agreements. Contact Rime to inquire about on-prem licensing.

Quickstart

Here is a basic example payload:

Output

json
{
  "text": "Hello! This is Rime speaking.",
  "speaker": "celeste",
  "lang": [
    "en",
    "es",
    "fr",
    "de"
  ]
}

Sending this request in Python:

Input

python
import json
import urllib.request

BASETEN_API_KEY = "your_api_key_here"

headers = {
    "Accept": "audio/wav",
    "Authorization": f"Bearer {BASETEN_API_KEY}",
    "Content-Type": "application/json"
}

payload = {
    "text": "Hello! This is Rime speaking.",
    "speaker": "celeste",
    "modelId": "mistv3"
}

data = json.dumps(payload).encode("utf-8")

BASETEN_ENDPOINT = "https://<endpoint>.api.baseten.co/environments/production/predict"

request = urllib.request.Request(
    BASETEN_ENDPOINT,
    data=data,
    headers=headers,
    method="POST"
)

with urllib.request.urlopen(request) as response:
    with open("output.wav", "wb") as f:
        while chunk := response.read(4096):
            f.write(chunk)

print("Audio saved to output.wav")
javascript
const RIME_API_KEY = "your_api_key_here";

const response = await fetch(
  "https://users.rime.ai/v1/rime-tts",
  {
    method: "POST",
    headers: {
      "Accept": "audio/wav",
      "Authorization": `Bearer ${RIME_API_KEY}`,
      "Content-Type": "application/json"
    },
    body: JSON.stringify({
      text: "Hello! This is Rime speaking.",
      speaker: "celeste",
      modelId: "mistv3"
    })
  }
);

const buffer = await response.arrayBuffer();

require("fs").writeFileSync(
  "output.wav",
  Buffer.from(buffer)
);

console.log("Audio saved to output.wav");

Output

json
{
  "audio": "base64 stream"
}

2. Payload Parameters

2.1 Required parameters

  • text (str): The text to convert to speech.

  • speaker (str): The voice to use. Browse all available voices at rime.ai/docs/voices.

  • modelId (str): Set to mistv3 for this model.

2.2 Optional parameters

ParameterTypeDefaultDescription
`samplingRate`number`24000`Output sampling rate in Hz
`timeScaleFactor`number`1.0`A value below 1 slows down the audio, a value above 1 speeds up the audio.
`pauseBetweenBrackets`boolean`false`Inserts pauses where brackets appear in text.
`phonemizeBetweenBrackets`boolean`false`Enables custom pronunciation for text inside curly brackets (e.g. `{r1Ym}`). See below.
`inlineSpeedAlpha`string" "Comma-separated list of speed values applied to words in square brackets. Values > 1.0 speed up speech, < 1.0 slow it down. Example: “This is [slow] and [fast]”, use “0.5, 3” to make “slow” slower and “fast” faster.

Implementation Notes

Custom Pronunciation

Mist v3's phoneme-first architecture gives you deterministic, highly controllable pronunciation;  critical for brand names, medical terms, and domain-specific vocabulary. To use custom pronunciation, wrap the word in curly brackets and set phonemizeBetweenBrackets to true:

Use the Pronunciation tool in the Rime dashboard to generate phonetic strings for any word.

Input

python
payload = {
    "text": "Welcome to {r1Ym} labs.",
    "speaker": "peak",
    "modelId": "mistv3",
    "phonemizeBetweenBrackets": True
}

Output

json
{
  "audio": "base64 stream"
}