README.md

---
license: apache-2.0
language:
  - en
tags:
  - text-to-speech
base_model:
  - sesame/csm-1b
pipeline_tag: text-to-speech
---

# CSM FP16 Safetensors

**2025/03/15** - This is the half-precision (FP16) Safetensors version of the 1B CSM variant which was [originally released in full-precision by Sesame](https://huggingface.co/sesame/csm_1b) on 2025/03/13.

---

CSM (Conversational Speech Model) is a speech generation model from [Sesame](https://www.sesame.com) that generates RVQ audio codes from text and audio inputs. The model architecture employs a [Llama](https://www.llama.com/) backbone and a smaller audio decoder that produces [Mimi](https://huggingface.co/kyutai/mimi) audio codes.

A fine-tuned variant of CSM powers the [interactive voice demo](https://www.sesame.com/voicedemo) shown in our [blog post](https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice).

A hosted [Hugging Face space](https://huggingface.co/spaces/sesame/csm-1b) is also available for testing audio generation.

## Conversion Statistics

Some statistics for the conversion from full-precision to half-precision:
- Original size: 6.22 GB
- Converted size: 3.11 GB
- Size reduction: 49.93%
- Max absolute difference: 0.000897
- Max relative difference: 0.229582
- Avg absolute difference: 0.000016

## Requirements

* A CUDA-compatible GPU
* The code has been tested on CUDA 12.4 and 12.6, but it may also work on other versions
* Simiarly, Python 3.10 is recommended, but newer versions may be fine
* For some audio operations, `ffmpeg` may be required

### Setup

```bash
git clone [email protected]:SesameAILabs/csm.git
cd csm
python3.10 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
pip install safetensors
```

### Windows Setup

The `triton` package cannot be installed in Windows. Instead use `pip install triton-windows`.

## Usage

Generate a sentence

```python
from huggingface_hub import hf_hub_download
from generator import Generator
from models import Model, ModelArgs
from safetensors.torch import load_file
import torchaudio
import torch

device = "cpu"
model_path = hf_hub_download(repo_id="thepushkarp/csm-1b-safetensors-fp16", filename="model.safetensors")

model_args = ModelArgs(
    backbone_flavor="llama-1B",
    decoder_flavor="llama-100M",
    text_vocab_size=128256,
    audio_vocab_size=2051,
    audio_num_codebooks=32,
)
model = Model(model_args).to(device=device, dtype=torch.float16)
loaded = load_file(model_path)
model.load_state_dict(loaded)

generator = Generator(model)
audio = generator.generate(
    text="Hello from Sesame.",
    speaker=0,
    context=[],
    max_audio_length_ms=10_000,
)

torchaudio.save("audio.wav", audio.unsqueeze(0).cpu(), generator.sample_rate)
```

CSM sounds best when provided with context. You can prompt or provide context to the model using a `Segment` for each speaker utterance.

```python
speakers = [0, 1, 0, 0]
transcripts = [
    "Hey how are you doing.",
    "Pretty good, pretty good.",
    "I'm great.",
    "So happy to be speaking to you.",
]
audio_paths = [
    "utterance_0.wav",
    "utterance_1.wav",
    "utterance_2.wav",
    "utterance_3.wav",
]

def load_audio(audio_path):
    audio_tensor, sample_rate = torchaudio.load(audio_path)
    audio_tensor = torchaudio.functional.resample(
        audio_tensor.squeeze(0), orig_freq=sample_rate, new_freq=generator.sample_rate
    )
    return audio_tensor

segments = [
    Segment(text=transcript, speaker=speaker, audio=load_audio(audio_path))
    for transcript, speaker, audio_path in zip(transcripts, speakers, audio_paths)
]
audio = generator.generate(
    text="Me too, this is some cool stuff huh?",
    speaker=1,
    context=segments,
    max_audio_length_ms=10_000,
)

torchaudio.save("audio.wav", audio.unsqueeze(0).cpu(), generator.sample_rate)
```

## FAQ

**Does this model come with any voices?**

The model open sourced here is a base generation model. It is capable of producing a variety of voices, but it has not been fine-tuned on any specific voice.

**Can I converse with the model?**

CSM is trained to be an audio generation model and not a general purpose multimodal LLM. It cannot generate text. We suggest using a separate LLM for text generation.

**Does it support other languages?**

The model has some capacity for non-English languages due to data contamination in the training data, but it likely won't do well.

## Misuse and abuse ⚠️

This project provides a high-quality speech generation model for research and educational purposes. While we encourage responsible and ethical use, we **explicitly prohibit** the following:

- **Impersonation or Fraud**: Do not use this model to generate speech that mimics real individuals without their explicit consent.
- **Misinformation or Deception**: Do not use this model to create deceptive or misleading content, such as fake news or fraudulent calls.
- **Illegal or Harmful Activities**: Do not use this model for any illegal, harmful, or malicious purposes.

By using this model, you agree to comply with all applicable laws and ethical guidelines. We are **not responsible** for any misuse, and we strongly condemn unethical applications of this technology.

---

## Authors
Johan Schalkwyk, Ankit Kumar, Dan Lyth, Sefik Emre Eskimez, Zack Hodari, Cinjon Resnick, Ramon Sanabria, Raven Jiang, and the Sesame team.