chatterbox-ONNX

<img width="800" alt="cb-big2" src="https://github.com/user-attachments/assets/bd8c5f03-e91d-4ee5-b680-57355da204d1" /> <h1 style="font-size: 32px">Chatterbox TTS</h1> <div style="display: flex; align-items: center; gap: 12px"> <a href="https://resemble-ai.github.io/chatterbox_demopage/"> <img src="https://img.shields.io/badge/listen-demo_samples-blue" alt="Listen to Demo Samples" /> </a> <a href="https://huggingface.co/spaces/ResembleAI/Chatterbox"> <img src="https://huggingface.co/datasets/huggingface/badges/resolve/main/open-in-hf-spaces-sm.svg" alt="Open in HF Spaces" /> </a> <a href="https://podonos.com/resembleai/chatterbox"> <img src="https://static-public.podonos.com/badges/insight-on-pdns-sm-dark.svg" alt="Insight on Podos" /> </a> </div> <div style="display: flex; align-items: center; gap: 8px;"> <img width="100" alt="resemble-logo-horizontal" src="https://github.com/user-attachments/assets/35cf756b-3506-4943-9c72-c05ddfa4e525" /> </div>

Chatterbox Resemble AI's production-grade open source TTS model. Chatterbox supports English out of the box. 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.

Chatterbox is provided in an exported ONNX format, enabling fast and portable inference with ONNX Runtime across platforms.

Key Details

• SoTA zeroshot English 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 (optional)
• Easy voice conversion script using onnxruntime
• Outperforms ElevenLabs

Tips

• General Use (TTS and Voice Agents):

  • The default settings (exaggeration=0.5, cfg=0.5) work well for most prompts.

• Expressive or Dramatic Speech:

  • Try increase exaggeration to around 0.7 or higher.
  • Higher exaggeration tends to speed up speech;

Usage

Link to GitHub ONNX Export and Inference script

# !pip install --upgrade onnxruntime==1.22.1 huggingface_hub==0.34.4 transformers==4.46.3 numpy==2.2.6 tqdm==4.67.1 librosa==0.11.0 soundfile==0.13.1 resemble-perth==1.0.1

import onnxruntime

from huggingface_hub import hf_hub_download
from transformers import AutoTokenizer

import numpy as np
from tqdm import tqdm
import librosa
import soundfile as sf

S3GEN_SR = 24000
START_SPEECH_TOKEN = 6561
STOP_SPEECH_TOKEN = 6562


class RepetitionPenaltyLogitsProcessor:
    def __init__(self, penalty: float):
        if not isinstance(penalty, float) or not (penalty > 0):
            raise ValueError(f"`penalty` must be a strictly positive float, but is {penalty}")
        self.penalty = penalty

    def __call__(self, input_ids: np.ndarray, scores: np.ndarray) -> np.ndarray:
        score = np.take_along_axis(scores, input_ids, axis=1)
        score = np.where(score < 0, score * self.penalty, score / self.penalty)
        scores_processed = scores.copy()
        np.put_along_axis(scores_processed, input_ids, score, axis=1)
        return scores_processed


def run_inference(
    text="The Lord of the Rings is the greatest work of literature.", 
    target_voice_path=None, 
    max_new_tokens = 256,
    exaggeration=0.5, 
    output_dir="converted", 
    output_file_name="output.wav",
    apply_watermark=True,
):

    model_id = "onnx-community/chatterbox-onnx"
    if not target_voice_path:
        target_voice_path = hf_hub_download(repo_id=model_id, filename="default_voice.wav", local_dir=output_dir)

    ## Load model
    speech_encoder_path = hf_hub_download(repo_id=model_id, filename="speech_encoder.onnx", local_dir=output_dir, subfolder='onnx')
    hf_hub_download(repo_id=model_id, filename="speech_encoder.onnx_data", local_dir=output_dir, subfolder='onnx')
    embed_tokens_path = hf_hub_download(repo_id=model_id, filename="embed_tokens.onnx", local_dir=output_dir, subfolder='onnx')
    hf_hub_download(repo_id=model_id, filename="embed_tokens.onnx_data", local_dir=output_dir, subfolder='onnx')
    conditional_decoder_path = hf_hub_download(repo_id=model_id, filename="conditional_decoder.onnx", local_dir=output_dir, subfolder='onnx')
    hf_hub_download(repo_id=model_id, filename="conditional_decoder.onnx_data", local_dir=output_dir, subfolder='onnx')
    language_model_path = hf_hub_download(repo_id=model_id, filename="language_model.onnx", local_dir=output_dir, subfolder='onnx')
    hf_hub_download(repo_id=model_id, filename="language_model.onnx_data", local_dir=output_dir, subfolder='onnx')

    # # Start inferense sessions
    speech_encoder_session = onnxruntime.InferenceSession(speech_encoder_path)
    embed_tokens_session = onnxruntime.InferenceSession(embed_tokens_path)
    llama_with_past_session = onnxruntime.InferenceSession(language_model_path)
    cond_decoder_session = onnxruntime.InferenceSession(conditional_decoder_path)

    def execute_text_to_audio_inference(text):
        print("Start inference script...")

        audio_values, _ = librosa.load(target_voice_path, sr=S3GEN_SR)
        audio_values = audio_values[np.newaxis, :].astype(np.float32)

        ## Prepare input
        tokenizer = AutoTokenizer.from_pretrained(model_id)
        input_ids = tokenizer(text, return_tensors="np")["input_ids"].astype(np.int64)

        position_ids = np.where(
            input_ids >= START_SPEECH_TOKEN,
            0,
            np.arange(input_ids.shape[1])[np.newaxis, :] - 1
        )

        ort_embed_tokens_inputs = {
            "input_ids": input_ids,
            "position_ids": position_ids,
            "exaggeration": np.array([exaggeration], dtype=np.float32)
        }

        ## Instantiate the logits processors.
        repetition_penalty = 1.2
        repetition_penalty_processor = RepetitionPenaltyLogitsProcessor(penalty=repetition_penalty)

        num_hidden_layers = 30
        num_key_value_heads = 16
        head_dim = 64

        generate_tokens = np.array([[START_SPEECH_TOKEN]], dtype=np.long)

        # ---- Generation Loop using kv_cache ----
        for i in tqdm(range(max_new_tokens), desc="Sampling", dynamic_ncols=True):

            inputs_embeds = embed_tokens_session.run(None, ort_embed_tokens_inputs)[0]
            if i == 0:
                ort_speech_encoder_input = {
                    "audio_values": audio_values,
                }
                cond_emb, prompt_token, ref_x_vector, prompt_feat = speech_encoder_session.run(None, ort_speech_encoder_input)
                inputs_embeds = np.concatenate((cond_emb, inputs_embeds), axis=1)

                ## Prepare llm inputs
                batch_size, seq_len, _ = inputs_embeds.shape
                past_key_values = {
                    f"past_key_values.{layer}.{kv}": np.zeros([batch_size, num_key_value_heads, 0, head_dim], dtype=np.float32)
                    for layer in range(num_hidden_layers)
                    for kv in ("key", "value")
                }
                attention_mask = np.ones((batch_size, seq_len), dtype=np.int64)

            logits, *present_key_values = llama_with_past_session.run(None, dict(
                inputs_embeds=inputs_embeds,
                attention_mask=attention_mask,
                **past_key_values,
            ))

            logits = logits[:, -1, :]
            next_token_logits = repetition_penalty_processor(generate_tokens, logits)

            next_token = np.argmax(next_token_logits, axis=-1, keepdims=True).astype(np.int64)
            generate_tokens = np.concatenate((generate_tokens, next_token), axis=-1)
            if (next_token.flatten() == STOP_SPEECH_TOKEN).all():
                break

            # Get embedding for the new token.
            position_ids = np.full(
                (input_ids.shape[0], 1),
                i + 1,
                dtype=np.int64,
            )
            ort_embed_tokens_inputs["input_ids"] = next_token
            ort_embed_tokens_inputs["position_ids"] = position_ids

            ## Update values for next generation loop
            attention_mask = np.concatenate([attention_mask, np.ones((batch_size, 1), dtype=np.int64)], axis=1)
            for j, key in enumerate(past_key_values):
                past_key_values[key] = present_key_values[j]

        speech_tokens = generate_tokens[:, 1:-1]
        speech_tokens = np.concatenate([prompt_token, speech_tokens], axis=1)
        return speech_tokens, ref_x_vector, prompt_feat

    speech_tokens, speaker_embeddings, speaker_features = execute_text_to_audio_inference(text)
    cond_incoder_input = {
        "speech_tokens": speech_tokens,
        "speaker_embeddings": speaker_embeddings,
        "speaker_features": speaker_features,
    }
    wav = cond_decoder_session.run(None, cond_incoder_input)[0]
    wav = np.squeeze(wav, axis=0)

    # Optional: Apply watermark
    if apply_watermark:
        import perth
        watermarker = perth.PerthImplicitWatermarker()
        wav = watermarker.apply_watermark(wav, sample_rate=S3GEN_SR)

    sf.write(output_file_name, wav, S3GEN_SR)
    print(f"{output_file_name} was successfully saved")

if __name__ == "__main__":
    run_inference(
        text="Ezreal and Jinx teamed up with Ahri, Yasuo, and Teemo to take down the enemy's Nexus in an epic late-game pentakill.",
        exaggeration=0.5,
        output_file_name="output.wav",
        apply_watermark=False,
    )

Acknowledgements

• Xenova
• Vladislav Bronzov
• Resemble AI

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.