resemble-ai / chatterbox

Generate expressive, natural speech. Features unique emotion control, instant voice cloning from short audio, and built-in watermarking.

Warm

Public
15.4K runs
$0.025 per thousand input characters
Weights
License

Iterate in playground

Run with an API

Playground API Examples README

Input

prompt

*string

Shift + Return to add a new line

We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.

Whether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.

If you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.

Whether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.

If you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.

Text to synthesize

audio_prompt

file

Path to the reference audio file (Optional)

exaggeration

number

(minimum: 0.25, maximum: 2)

Exaggeration (Neutral = 0.5, extreme values can be unstable)

Default: 0.5

cfg_weight

number

(minimum: 0.2, maximum: 1)

CFG/Pace weight

Default: 0.5

temperature

number

(minimum: 0.05, maximum: 5)

Temperature

Default: 0.8

seed

integer

Seed (0 for random)

Default: 0

Run this model in Node.js with one line of code:

npx create-replicate --model=resemble-ai/chatterbox

or set up a project from scratch

Install Replicate’s Node.js client library:

npm install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client:

import Replicate from "replicate";
import fs from "node:fs";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run resemble-ai/chatterbox using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const input = {
  seed: 0,
  prompt: "We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.\n\nWhether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.\n\nIf you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.",
  cfg_weight: 0.5,
  temperature: 0.8,
  exaggeration: 0.5
};

const output = await replicate.run("resemble-ai/chatterbox", { input });

// To access the file URL:
console.log(output.url()); //=> "http://example.com"

// To write the file to disk:
fs.writeFile("my-image.png", output);

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library:

pip install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client:

import replicate

Run resemble-ai/chatterbox using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

output = replicate.run(
    "resemble-ai/chatterbox",
    input={
        "seed": 0,
        "prompt": "We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.\n\nWhether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.\n\nIf you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.",
        "cfg_weight": 0.5,
        "temperature": 0.8,
        "exaggeration": 0.5
    }
)
print(output)

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run resemble-ai/chatterbox using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "input": {
      "seed": 0,
      "prompt": "We\'re excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.\\n\\nWhether you\'re working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It\'s also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.\\n\\nIf you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.",
      "cfg_weight": 0.5,
      "temperature": 0.8,
      "exaggeration": 0.5
    }
  }' \
  https://api.replicate.com/v1/models/resemble-ai/chatterbox/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

Output

Video Player is loading.

Current Time 00:00:000

Duration 00:00:000

Loaded: 0%

Stream Type LIVE

Remaining Time 00:00:000

{
  "completed_at": "2025-06-12T15:50:17.756333Z",
  "created_at": "2025-06-12T15:50:04.982000Z",
  "data_removed": false,
  "error": null,
  "id": "g7r2drc7psrmc0cqckasg1z4ec",
  "input": {
    "seed": 0,
    "prompt": "We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.\n\nWhether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.\n\nIf you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.",
    "cfg_weight": 0.5,
    "temperature": 0.8,
    "exaggeration": 0.5
  },
  "logs": "Using seed: 35127\nPrompt: We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.\nWhether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.\nIf you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.\n/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/cog/server/scope.py:22: ExperimentalFeatureWarning: current_scope is an experimental internal function. It may change or be removed without warning.\nwarnings.warn(\n/root/.pyenv/versions/3.11.10/lib/python3.11/contextlib.py:105: FutureWarning: `torch.backends.cuda.sdp_kernel()` is deprecated. In the future, this context manager will be removed. Please see `torch.nn.attention.sdpa_kernel()` for the new context manager, with updated signature.\nself.gen = func(*args, **kwds)\nSampling:   0%|          | 0/1000 [00:00<?, ?it/s]\nSampling:   1%|          | 6/1000 [00:00<00:18, 55.18it/s]\nSampling:   2%|▏         | 15/1000 [00:00<00:13, 74.76it/s]\nSampling:   2%|▏         | 24/1000 [00:00<00:12, 81.19it/s]\nSampling:   3%|▎         | 33/1000 [00:00<00:11, 83.63it/s]\nSampling:   4%|▍         | 42/1000 [00:00<00:11, 85.73it/s]\nSampling:   5%|▌         | 51/1000 [00:00<00:10, 87.05it/s]\nSampling:   6%|▌         | 61/1000 [00:00<00:10, 88.09it/s]\nSampling:   7%|▋         | 70/1000 [00:00<00:10, 88.63it/s]\nSampling:   8%|▊         | 79/1000 [00:00<00:10, 88.90it/s]\nSampling:   9%|▉         | 88/1000 [00:01<00:10, 89.23it/s]\nSampling:  10%|▉         | 98/1000 [00:01<00:10, 89.51it/s]\nSampling:  11%|█         | 107/1000 [00:01<00:10, 88.86it/s]\nSampling:  12%|█▏        | 116/1000 [00:01<00:09, 88.84it/s]\nSampling:  12%|█▎        | 125/1000 [00:01<00:10, 85.84it/s]\nSampling:  13%|█▎        | 134/1000 [00:01<00:09, 86.92it/s]\nSampling:  14%|█▍        | 143/1000 [00:01<00:09, 87.78it/s]\nSampling:  15%|█▌        | 153/1000 [00:01<00:09, 88.63it/s]\nSampling:  16%|█▋        | 163/1000 [00:01<00:09, 89.36it/s]\nSampling:  17%|█▋        | 172/1000 [00:01<00:09, 89.25it/s]\nSampling:  18%|█▊        | 181/1000 [00:02<00:09, 89.10it/s]\nSampling:  19%|█▉        | 190/1000 [00:02<00:09, 88.12it/s]\nSampling:  20%|█▉        | 199/1000 [00:02<00:09, 88.61it/s]\nSampling:  21%|██        | 208/1000 [00:02<00:08, 88.88it/s]\nSampling:  22%|██▏       | 217/1000 [00:02<00:08, 88.95it/s]\nSampling:  23%|██▎       | 226/1000 [00:02<00:09, 85.92it/s]\nSampling:  24%|██▎       | 235/1000 [00:02<00:08, 85.71it/s]\nSampling:  24%|██▍       | 245/1000 [00:02<00:08, 87.17it/s]\nSampling:  26%|██▌       | 255/1000 [00:02<00:08, 88.27it/s]\nSampling:  26%|██▋       | 265/1000 [00:03<00:08, 89.05it/s]\nSampling:  27%|██▋       | 274/1000 [00:03<00:08, 89.05it/s]\nSampling:  28%|██▊       | 284/1000 [00:03<00:07, 89.58it/s]\nSampling:  29%|██▉       | 294/1000 [00:03<00:07, 89.95it/s]\nSampling:  30%|███       | 303/1000 [00:03<00:07, 89.50it/s]\nSampling:  31%|███       | 312/1000 [00:03<00:07, 89.21it/s]\nSampling:  32%|███▏      | 322/1000 [00:03<00:07, 88.06it/s]\nSampling:  33%|███▎      | 331/1000 [00:03<00:08, 82.01it/s]\nSampling:  34%|███▍      | 340/1000 [00:03<00:08, 76.62it/s]\nSampling:  35%|███▍      | 349/1000 [00:04<00:08, 79.28it/s]\nSampling:  36%|███▌      | 358/1000 [00:04<00:07, 82.15it/s]\nSampling:  37%|███▋      | 368/1000 [00:04<00:07, 84.76it/s]\nSampling:  38%|███▊      | 377/1000 [00:04<00:07, 85.59it/s]\nSampling:  39%|███▊      | 386/1000 [00:04<00:07, 86.24it/s]\nSampling:  40%|███▉      | 395/1000 [00:04<00:06, 87.28it/s]\nSampling:  40%|████      | 404/1000 [00:04<00:06, 87.98it/s]\nSampling:  41%|████▏     | 414/1000 [00:04<00:06, 88.65it/s]\nSampling:  42%|████▏     | 423/1000 [00:04<00:06, 88.67it/s]\nSampling:  43%|████▎     | 432/1000 [00:04<00:06, 87.79it/s]\nSampling:  44%|████▍     | 441/1000 [00:05<00:06, 81.04it/s]\nSampling:  45%|████▌     | 450/1000 [00:05<00:06, 83.47it/s]\nSampling:  46%|████▌     | 459/1000 [00:05<00:06, 84.75it/s]\nSampling:  47%|████▋     | 468/1000 [00:05<00:06, 85.95it/s]\nSampling:  48%|████▊     | 477/1000 [00:05<00:06, 86.40it/s]\nSampling:  49%|████▊     | 486/1000 [00:05<00:05, 86.96it/s]\nSampling:  50%|████▉     | 495/1000 [00:05<00:06, 84.11it/s]\nSampling:  50%|█████     | 504/1000 [00:05<00:06, 79.72it/s]\nSampling:  51%|█████▏    | 513/1000 [00:05<00:06, 75.46it/s]\nSampling:  52%|█████▏    | 521/1000 [00:06<00:06, 74.52it/s]\nSampling:  53%|█████▎    | 529/1000 [00:06<00:06, 74.46it/s]\nSampling:  54%|█████▍    | 538/1000 [00:06<00:05, 77.79it/s]\nSampling:  55%|█████▍    | 547/1000 [00:06<00:05, 79.97it/s]\nSampling:  56%|█████▌    | 556/1000 [00:06<00:05, 81.23it/s]\nSampling:  56%|█████▋    | 565/1000 [00:06<00:05, 82.66it/s]\nSampling:  57%|█████▋    | 574/1000 [00:06<00:05, 83.38it/s]\nSampling:  58%|█████▊    | 583/1000 [00:06<00:04, 84.27it/s]\nSampling:  59%|█████▉    | 592/1000 [00:06<00:04, 85.06it/s]\nSampling:  60%|██████    | 601/1000 [00:07<00:04, 84.42it/s]\nSampling:  61%|██████    | 610/1000 [00:07<00:04, 84.80it/s]\nSampling:  62%|██████▏   | 619/1000 [00:07<00:04, 84.83it/s]\nSampling:  63%|██████▎   | 628/1000 [00:07<00:04, 84.92it/s]\nSampling:  64%|██████▎   | 637/1000 [00:07<00:04, 84.61it/s]\nSampling:  65%|██████▍   | 646/1000 [00:07<00:04, 85.21it/s]\nSampling:  66%|██████▌   | 655/1000 [00:07<00:04, 85.73it/s]\nSampling:  66%|██████▋   | 664/1000 [00:07<00:03, 84.85it/s]\nSampling:  67%|██████▋   | 673/1000 [00:07<00:03, 85.35it/s]\nSampling:  68%|██████▊   | 682/1000 [00:08<00:03, 85.06it/s]\nSampling:  69%|██████▉   | 691/1000 [00:08<00:03, 85.40it/s]\nSampling:  70%|███████   | 700/1000 [00:08<00:03, 85.67it/s]\nSampling:  71%|███████   | 709/1000 [00:08<00:03, 86.28it/s]\nSampling:  72%|███████▏  | 718/1000 [00:08<00:03, 87.08it/s]\nSampling:  73%|███████▎  | 727/1000 [00:08<00:03, 87.74it/s]\nSampling:  74%|███████▎  | 736/1000 [00:08<00:03, 87.56it/s]\nSampling:  74%|███████▍  | 745/1000 [00:08<00:02, 87.25it/s]\nSampling:  75%|███████▌  | 754/1000 [00:08<00:02, 86.45it/s]\nSampling:  76%|███████▋  | 763/1000 [00:08<00:02, 85.54it/s]\nSampling:  77%|███████▋  | 772/1000 [00:09<00:02, 85.73it/s]\nSampling:  78%|███████▊  | 781/1000 [00:09<00:02, 86.06it/s]\nSampling:  79%|███████▉  | 790/1000 [00:09<00:02, 74.94it/s]\nSampling:  80%|███████▉  | 799/1000 [00:09<00:02, 76.75it/s]\nSampling:  81%|████████  | 808/1000 [00:09<00:02, 79.31it/s]\nSampling:  82%|████████▏ | 817/1000 [00:09<00:02, 81.78it/s]\nSampling:  83%|████████▎ | 826/1000 [00:09<00:02, 83.09it/s]\nSampling:  84%|████████▎ | 835/1000 [00:09<00:01, 83.62it/s]\nSampling:  84%|████████▍ | 844/1000 [00:09<00:01, 83.99it/s]\nSampling:  85%|████████▌ | 853/1000 [00:10<00:01, 84.38it/s]\nSampling:  86%|████████▌ | 862/1000 [00:10<00:01, 84.34it/s]\nSampling:  87%|████████▋ | 871/1000 [00:10<00:01, 80.87it/s]\nSampling:  88%|████████▊ | 880/1000 [00:10<00:01, 81.79it/s]\nSampling:  89%|████████▉ | 889/1000 [00:10<00:01, 81.93it/s]\nSampling:  90%|████████▉ | 898/1000 [00:10<00:01, 83.45it/s]\nSampling:  91%|█████████ | 907/1000 [00:10<00:01, 84.63it/s]\nSampling:  92%|█████████▏| 916/1000 [00:10<00:00, 85.65it/s]\nSampling:  92%|█████████▎| 925/1000 [00:10<00:00, 86.04it/s]\nSampling:  93%|█████████▎| 934/1000 [00:11<00:00, 84.32it/s]\nSampling:  94%|█████████▍| 943/1000 [00:11<00:00, 84.61it/s]\nSampling:  95%|█████████▌| 952/1000 [00:11<00:00, 85.10it/s]\nSampling:  96%|█████████▌| 961/1000 [00:11<00:00, 84.62it/s]\nSampling:  97%|█████████▋| 970/1000 [00:11<00:00, 85.25it/s]\nSampling:  98%|█████████▊| 979/1000 [00:11<00:00, 85.66it/s]\nSampling:  98%|█████████▊| 981/1000 [00:11<00:00, 84.85it/s]\nInput character count: 809\nCharacters per second: 63.55\nTotal time: 12.73s",
  "metrics": {
    "input_token_count": 161,
    "tokens_per_second": 0.07833769183347587,
    "output_token_count": 1,
    "predict_time": 12.765277608,
    "total_time": 12.774333,
    "time_to_first_token": 0.009085425000000001
  },
  "output": "https://replicate.delivery/xezq/xx8Wscg0ve0gfko6krsG77SUuy7ZsQINBrJuRgq6GfCzJVspA/output.wav",
  "started_at": "2025-06-12T15:50:04.991055Z",
  "status": "succeeded",
  "urls": {
    "stream": "https://stream.replicate.com/v1/files/bcwr-alelxyvdrsmhelqnady6t6w6p7mnhz4vsvyaa2zviel5mupyrbaa",
    "get": "https://api.replicate.com/v1/predictions/g7r2drc7psrmc0cqckasg1z4ec",
    "cancel": "https://api.replicate.com/v1/predictions/g7r2drc7psrmc0cqckasg1z4ec/cancel"
  },
  "version": "hidden"
}

Generated in

12.8 seconds

Input tokens

160

Output tokens

Tokens per second

0.08 tokens / second

Time to first token

9 milliseconds

Tweak itReport View full prediction

Using seed: 35127
Prompt: We're excited to introduce Chatterbox, our first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.
Whether you're working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It's also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.
If you like the model but need to scale or finetune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.
/root/.pyenv/versions/3.11.10/lib/python3.11/site-packages/cog/server/scope.py:22: ExperimentalFeatureWarning: current_scope is an experimental internal function. It may change or be removed without warning.
warnings.warn(
/root/.pyenv/versions/3.11.10/lib/python3.11/contextlib.py:105: FutureWarning: `torch.backends.cuda.sdp_kernel()` is deprecated. In the future, this context manager will be removed. Please see `torch.nn.attention.sdpa_kernel()` for the new context manager, with updated signature.
self.gen = func(*args, **kwds)
Sampling:   0%|          | 0/1000 [00:00<?, ?it/s]
Sampling:   1%|          | 6/1000 [00:00<00:18, 55.18it/s]
Sampling:   2%|▏         | 15/1000 [00:00<00:13, 74.76it/s]
Sampling:   2%|▏         | 24/1000 [00:00<00:12, 81.19it/s]
Sampling:   3%|▎         | 33/1000 [00:00<00:11, 83.63it/s]
Sampling:   4%|▍         | 42/1000 [00:00<00:11, 85.73it/s]
Sampling:   5%|▌         | 51/1000 [00:00<00:10, 87.05it/s]
Sampling:   6%|▌         | 61/1000 [00:00<00:10, 88.09it/s]
Sampling:   7%|▋         | 70/1000 [00:00<00:10, 88.63it/s]
Sampling:   8%|▊         | 79/1000 [00:00<00:10, 88.90it/s]
Sampling:   9%|▉         | 88/1000 [00:01<00:10, 89.23it/s]
Sampling:  10%|▉         | 98/1000 [00:01<00:10, 89.51it/s]
Sampling:  11%|█         | 107/1000 [00:01<00:10, 88.86it/s]
Sampling:  12%|█▏        | 116/1000 [00:01<00:09, 88.84it/s]
Sampling:  12%|█▎        | 125/1000 [00:01<00:10, 85.84it/s]
Sampling:  13%|█▎        | 134/1000 [00:01<00:09, 86.92it/s]
Sampling:  14%|█▍        | 143/1000 [00:01<00:09, 87.78it/s]
Sampling:  15%|█▌        | 153/1000 [00:01<00:09, 88.63it/s]
Sampling:  16%|█▋        | 163/1000 [00:01<00:09, 89.36it/s]
Sampling:  17%|█▋        | 172/1000 [00:01<00:09, 89.25it/s]
Sampling:  18%|█▊        | 181/1000 [00:02<00:09, 89.10it/s]
Sampling:  19%|█▉        | 190/1000 [00:02<00:09, 88.12it/s]
Sampling:  20%|█▉        | 199/1000 [00:02<00:09, 88.61it/s]
Sampling:  21%|██        | 208/1000 [00:02<00:08, 88.88it/s]
Sampling:  22%|██▏       | 217/1000 [00:02<00:08, 88.95it/s]
Sampling:  23%|██▎       | 226/1000 [00:02<00:09, 85.92it/s]
Sampling:  24%|██▎       | 235/1000 [00:02<00:08, 85.71it/s]
Sampling:  24%|██▍       | 245/1000 [00:02<00:08, 87.17it/s]
Sampling:  26%|██▌       | 255/1000 [00:02<00:08, 88.27it/s]
Sampling:  26%|██▋       | 265/1000 [00:03<00:08, 89.05it/s]
Sampling:  27%|██▋       | 274/1000 [00:03<00:08, 89.05it/s]
Sampling:  28%|██▊       | 284/1000 [00:03<00:07, 89.58it/s]
Sampling:  29%|██▉       | 294/1000 [00:03<00:07, 89.95it/s]
Sampling:  30%|███       | 303/1000 [00:03<00:07, 89.50it/s]
Sampling:  31%|███       | 312/1000 [00:03<00:07, 89.21it/s]
Sampling:  32%|███▏      | 322/1000 [00:03<00:07, 88.06it/s]
Sampling:  33%|███▎      | 331/1000 [00:03<00:08, 82.01it/s]
Sampling:  34%|███▍      | 340/1000 [00:03<00:08, 76.62it/s]
Sampling:  35%|███▍      | 349/1000 [00:04<00:08, 79.28it/s]
Sampling:  36%|███▌      | 358/1000 [00:04<00:07, 82.15it/s]
Sampling:  37%|███▋      | 368/1000 [00:04<00:07, 84.76it/s]
Sampling:  38%|███▊      | 377/1000 [00:04<00:07, 85.59it/s]
Sampling:  39%|███▊      | 386/1000 [00:04<00:07, 86.24it/s]
Sampling:  40%|███▉      | 395/1000 [00:04<00:06, 87.28it/s]
Sampling:  40%|████      | 404/1000 [00:04<00:06, 87.98it/s]
Sampling:  41%|████▏     | 414/1000 [00:04<00:06, 88.65it/s]
Sampling:  42%|████▏     | 423/1000 [00:04<00:06, 88.67it/s]
Sampling:  43%|████▎     | 432/1000 [00:04<00:06, 87.79it/s]
Sampling:  44%|████▍     | 441/1000 [00:05<00:06, 81.04it/s]
Sampling:  45%|████▌     | 450/1000 [00:05<00:06, 83.47it/s]
Sampling:  46%|████▌     | 459/1000 [00:05<00:06, 84.75it/s]
Sampling:  47%|████▋     | 468/1000 [00:05<00:06, 85.95it/s]
Sampling:  48%|████▊     | 477/1000 [00:05<00:06, 86.40it/s]
Sampling:  49%|████▊     | 486/1000 [00:05<00:05, 86.96it/s]
Sampling:  50%|████▉     | 495/1000 [00:05<00:06, 84.11it/s]
Sampling:  50%|█████     | 504/1000 [00:05<00:06, 79.72it/s]
Sampling:  51%|█████▏    | 513/1000 [00:05<00:06, 75.46it/s]
Sampling:  52%|█████▏    | 521/1000 [00:06<00:06, 74.52it/s]
Sampling:  53%|█████▎    | 529/1000 [00:06<00:06, 74.46it/s]
Sampling:  54%|█████▍    | 538/1000 [00:06<00:05, 77.79it/s]
Sampling:  55%|█████▍    | 547/1000 [00:06<00:05, 79.97it/s]
Sampling:  56%|█████▌    | 556/1000 [00:06<00:05, 81.23it/s]
Sampling:  56%|█████▋    | 565/1000 [00:06<00:05, 82.66it/s]
Sampling:  57%|█████▋    | 574/1000 [00:06<00:05, 83.38it/s]
Sampling:  58%|█████▊    | 583/1000 [00:06<00:04, 84.27it/s]
Sampling:  59%|█████▉    | 592/1000 [00:06<00:04, 85.06it/s]
Sampling:  60%|██████    | 601/1000 [00:07<00:04, 84.42it/s]
Sampling:  61%|██████    | 610/1000 [00:07<00:04, 84.80it/s]
Sampling:  62%|██████▏   | 619/1000 [00:07<00:04, 84.83it/s]
Sampling:  63%|██████▎   | 628/1000 [00:07<00:04, 84.92it/s]
Sampling:  64%|██████▎   | 637/1000 [00:07<00:04, 84.61it/s]
Sampling:  65%|██████▍   | 646/1000 [00:07<00:04, 85.21it/s]
Sampling:  66%|██████▌   | 655/1000 [00:07<00:04, 85.73it/s]
Sampling:  66%|██████▋   | 664/1000 [00:07<00:03, 84.85it/s]
Sampling:  67%|██████▋   | 673/1000 [00:07<00:03, 85.35it/s]
Sampling:  68%|██████▊   | 682/1000 [00:08<00:03, 85.06it/s]
Sampling:  69%|██████▉   | 691/1000 [00:08<00:03, 85.40it/s]
Sampling:  70%|███████   | 700/1000 [00:08<00:03, 85.67it/s]
Sampling:  71%|███████   | 709/1000 [00:08<00:03, 86.28it/s]
Sampling:  72%|███████▏  | 718/1000 [00:08<00:03, 87.08it/s]
Sampling:  73%|███████▎  | 727/1000 [00:08<00:03, 87.74it/s]
Sampling:  74%|███████▎  | 736/1000 [00:08<00:03, 87.56it/s]
Sampling:  74%|███████▍  | 745/1000 [00:08<00:02, 87.25it/s]
Sampling:  75%|███████▌  | 754/1000 [00:08<00:02, 86.45it/s]
Sampling:  76%|███████▋  | 763/1000 [00:08<00:02, 85.54it/s]
Sampling:  77%|███████▋  | 772/1000 [00:09<00:02, 85.73it/s]
Sampling:  78%|███████▊  | 781/1000 [00:09<00:02, 86.06it/s]
Sampling:  79%|███████▉  | 790/1000 [00:09<00:02, 74.94it/s]
Sampling:  80%|███████▉  | 799/1000 [00:09<00:02, 76.75it/s]
Sampling:  81%|████████  | 808/1000 [00:09<00:02, 79.31it/s]
Sampling:  82%|████████▏ | 817/1000 [00:09<00:02, 81.78it/s]
Sampling:  83%|████████▎ | 826/1000 [00:09<00:02, 83.09it/s]
Sampling:  84%|████████▎ | 835/1000 [00:09<00:01, 83.62it/s]
Sampling:  84%|████████▍ | 844/1000 [00:09<00:01, 83.99it/s]
Sampling:  85%|████████▌ | 853/1000 [00:10<00:01, 84.38it/s]
Sampling:  86%|████████▌ | 862/1000 [00:10<00:01, 84.34it/s]
Sampling:  87%|████████▋ | 871/1000 [00:10<00:01, 80.87it/s]
Sampling:  88%|████████▊ | 880/1000 [00:10<00:01, 81.79it/s]
Sampling:  89%|████████▉ | 889/1000 [00:10<00:01, 81.93it/s]
Sampling:  90%|████████▉ | 898/1000 [00:10<00:01, 83.45it/s]
Sampling:  91%|█████████ | 907/1000 [00:10<00:01, 84.63it/s]
Sampling:  92%|█████████▏| 916/1000 [00:10<00:00, 85.65it/s]
Sampling:  92%|█████████▎| 925/1000 [00:10<00:00, 86.04it/s]
Sampling:  93%|█████████▎| 934/1000 [00:11<00:00, 84.32it/s]
Sampling:  94%|█████████▍| 943/1000 [00:11<00:00, 84.61it/s]
Sampling:  95%|█████████▌| 952/1000 [00:11<00:00, 85.10it/s]
Sampling:  96%|█████████▌| 961/1000 [00:11<00:00, 84.62it/s]
Sampling:  97%|█████████▋| 970/1000 [00:11<00:00, 85.25it/s]
Sampling:  98%|█████████▊| 979/1000 [00:11<00:00, 85.66it/s]
Sampling:  98%|█████████▊| 981/1000 [00:11<00:00, 84.85it/s]
Input character count: 809
Characters per second: 63.55
Total time: 12.73s

Pricing

Model pricing for resemble-ai/chatterbox. Looking for volume pricing? Get in touch.

$0.025

per thousand input characters

or 40,000 characters for $1

Official models are always on, maintained, and have predictable pricing. Learn more.

Check out our docs for more information about how pricing works on Replicate.

Readme

cb-big2

Chatterbox TTS

We’re excited to introduce Chatterbox, Resemble AI’s first production-grade open source TTS model. Licensed under MIT, Chatterbox has been benchmarked against leading closed-source systems like ElevenLabs, and is consistently preferred in side-by-side evaluations.

Whether you’re working on memes, videos, games, or AI agents, Chatterbox brings your content to life. It’s also the first open source TTS model to support emotion exaggeration control, a powerful feature that makes your voices stand out. Try it now on our Hugging Face Gradio app.

If you like the model but need to scale or tune it for higher accuracy, check out our competitively priced TTS service (link). It delivers reliable performance with ultra-low latency of sub 200ms—ideal for production use in agents, applications, or interactive media.

Key Details

SoTA zeroshot TTS
0.5B Llama backbone
Unique exaggeration/intensity control
Ultra-stable with alignment-informed inference
Trained on 0.5M hours of cleaned data
Watermarked outputs
Easy voice conversion script
Outperforms ElevenLabs

Tips

General Use (TTS and Voice Agents):
The default settings (exaggeration=0.5, cfg=0.5) work well for most prompts.
If the reference speaker has a fast speaking style, lowering cfg to around 0.3 can improve pacing.
Expressive or Dramatic Speech:
Try lower cfg values (e.g. ~0.3) and increase exaggeration to around 0.7 or higher.
Higher exaggeration tends to speed up speech; reducing cfg helps compensate with slower, more deliberate pacing.

Acknowledgements

Built-in PerTh Watermarking for Responsible AI

Every audio file generated by Chatterbox includes Resemble AI’s Perth (Perceptual Threshold) Watermarker - imperceptible neural watermarks that survive MP3 compression, audio editing, and common manipulations while maintaining nearly 100% detection accuracy.

Disclaimer

Don’t use this model to do bad things. Prompts are sourced from freely available data on the internet.