README.md

---
library_name: transformers
tags:
- text-to-speech
- annotation
license: apache-2.0
language:
- en
- fr
- es
- pt
- pl
- de
- nl
- it
pipeline_tag: text-to-speech
inference: false
datasets:
- facebook/multilingual_librispeech
- parler-tts/libritts_r_filtered
- parler-tts/libritts-r-filtered-speaker-descriptions
- parler-tts/mls_eng
- parler-tts/mls-eng-speaker-descriptions
- ylacombe/mls-annotated
- ylacombe/cml-tts-filtered-annotated
- PHBJT/cml-tts-filtered
---
  
<img src="https://huggingface.co/datasets/parler-tts/images/resolve/main/thumbnail.png" alt="Parler Logo" width="800" style="margin-left:'auto' margin-right:'auto' display:'block'"/>


# Parler-TTS Mini Multilingual v1.1

<a target="_blank" href="https://huggingface.co/spaces/PHBJT/multi_parler_tts">
  <img src="https://huggingface.co/datasets/huggingface/badges/raw/main/open-in-hf-spaces-sm.svg" alt="Open in HuggingFace"/>
</a>

**Parler-TTS Mini Multilingual v1.1** is a multilingual extension of [Parler-TTS Mini](https://huggingface.co/parler-tts/parler-tts-mini-v1.1).

🚨 As compared to [Mini Multilingual v1](https://huggingface.co/parler-tts/parler-tts-mini-multilingual), this version was trained with some consistent speaker names and with better format for descriptions. 🚨

It is a fine-tuned version, trained on a [cleaned version](https://huggingface.co/datasets/PHBJT/cml-tts-filtered) of [CML-TTS](https://huggingface.co/datasets/ylacombe/cml-tts) and on the non-English version of [Multilingual LibriSpeech](https://huggingface.co/datasets/facebook/multilingual_librispeech).
In all, this represents some 9,200 hours of non-English data. To retain English capabilities, we also added back the [LibriTTS-R English dataset](https://huggingface.co/datasets/parler-tts/libritts_r_filtered), some 580h of high-quality English data.

**Parler-TTS Mini Multilingual** can speak in 8 European languages: English, French, Spanish, Portuguese, Polish, German, Italian and Dutch. 

Thanks to its **better prompt tokenizer**, it can easily be extended to other languages. This tokenizer has a larger vocabulary and handles byte fallback, which simplifies multilingual training.

🚨 This work is the result of a collaboration between the **HuggingFace audio team** and the **[Quantum Squadra](https://quantumsquadra.com/) team**. The **[AI4Bharat](https://ai4bharat.iitm.ac.in/) team** also provided advice and assistance in improving tokenization. 🚨


## 📖 Quick Index
* [👨‍💻 Installation](#👨‍💻-installation)
* [🎲 Using a random voice](#🎲-random-voice)
* [🎯 Using a specific speaker](#🎯-using-a-specific-speaker)
* [Motivation](#motivation)
* [Optimizing inference](https://github.com/huggingface/parler-tts/blob/main/INFERENCE.md)

## 🛠️ Usage

🚨Unlike previous versions of Parler-TTS, here we use two tokenizers - one for the prompt and one for the description.🚨

### 👨‍💻 Installation

Using Parler-TTS is as simple as "bonjour". Simply install the library once:

```sh
pip install git+https://github.com/huggingface/parler-tts.git
```

### 🎲 Random voice

**Parler-TTS Mini Multilingual** has been trained to generate speech with features that can be controlled with a simple text prompt, for example:

```py
import torch
from parler_tts import ParlerTTSForConditionalGeneration
from transformers import AutoTokenizer
import soundfile as sf

device = "cuda:0" if torch.cuda.is_available() else "cpu"

model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-mini-multilingual-v1.1").to(device)
tokenizer = AutoTokenizer.from_pretrained("parler-tts/parler-tts-mini-multilingual-v1.1")
description_tokenizer = AutoTokenizer.from_pretrained(model.config.text_encoder._name_or_path)

prompt = "Salut toi, comment vas-tu aujourd'hui?"
description = "A female speaker delivers a slightly expressive and animated speech with a moderate speed and pitch. The recording is of very high quality, with the speaker's voice sounding clear and very close up."

input_ids = description_tokenizer(description, return_tensors="pt").input_ids.to(device)
prompt_input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)

generation = model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids)
audio_arr = generation.cpu().numpy().squeeze()
sf.write("parler_tts_out.wav", audio_arr, model.config.sampling_rate)
```

### 🎯 Using a specific speaker

To ensure speaker consistency across generations, this checkpoint was also trained on 16 speakers, characterized by name (e.g. Daniel, Christine, Richard, Nicole, ...).

To take advantage of this, simply adapt your text description to specify which speaker to use: `Daniel's voice is monotone yet slightly fast in delivery, with a very close recording that almost has no background noise.`

```py
import torch
from parler_tts import ParlerTTSForConditionalGeneration
from transformers import AutoTokenizer
import soundfile as sf

device = "cuda:0" if torch.cuda.is_available() else "cpu"

model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-mini-multilingual-v1.1").to(device)
tokenizer = AutoTokenizer.from_pretrained("parler-tts/parler-tts-mini-multilingual-v1.1")
description_tokenizer = AutoTokenizer.from_pretrained(model.config.text_encoder._name_or_path)

prompt = "Salut toi, comment vas-tu aujourd'hui?"
description = "Daniel's voice is monotone yet slightly fast in delivery, with a very close recording that almost has no background noise."

input_ids = description_tokenizer(description, return_tensors="pt").input_ids.to(device)
prompt_input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)

generation = model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids)
audio_arr = generation.cpu().numpy().squeeze()
sf.write("parler_tts_out.wav", audio_arr, model.config.sampling_rate)
```

You can choose a speaker from this list: 

| Language   | Speaker Name | Number of occurrences it was trained on |
|------------|--------------|-----------------------------------------|
| Dutch      | Mark         | 460066                                  |
|            | Jessica      | 4438                                    |
|            | Michelle     | 83                                      |
| French     | Daniel       | 10719                                   |
|            | Michelle     | 19                                      |
|            | Christine    | 20187                                   |
|            | Megan        | 695                                     |
| German     | Nicole       | 53964                                   |
|            | Christopher  | 1671                                    |
|            | Megan        | 41                                      |
|            | Michelle     | 12693                                   |
| Italian    | Julia        | 2616                                    |
|            | Richard      | 9640                                    |
|            | Megan        | 4                                       |
| Polish     | Alex         | 25849                                   |
|            | Natalie      | 9384                                    |
| Portuguese | Sophia       | 34182                                   |
|            | Nicholas     | 4411                                    |
| Spanish    | Steven       | 74099                                   |
|            | Olivia       | 48489                                   |
|            | Megan        | 12                                      |

**Tips**:
* We've set up an [inference guide](https://github.com/huggingface/parler-tts/blob/main/INFERENCE.md) to make generation faster. Think SDPA, torch.compile, batching and streaming!
* Include the term "very clear audio" to generate the highest quality audio, and "very noisy audio" for high levels of background noise
* Punctuation can be used to control the prosody of the generations, e.g. use commas to add small breaks in speech
* The remaining speech features (gender, speaking rate, pitch and reverberation) can be controlled directly through the prompt

## Motivation

Parler-TTS is a reproduction of work from the paper [Natural language guidance of high-fidelity text-to-speech with synthetic annotations](https://www.text-description-to-speech.com) by Dan Lyth and Simon King, from Stability AI and Edinburgh University respectively. 

Contrarily to other TTS models, Parler-TTS is a **fully open-source** release. All of the datasets, pre-processing, training code and weights are released publicly under permissive license, enabling the community to build on our work and develop their own powerful TTS models.
Parler-TTS was released alongside:
* [The Parler-TTS repository](https://github.com/huggingface/parler-tts) - you can train and fine-tuned your own version of the model.
* [The Data-Speech repository](https://github.com/huggingface/dataspeech) - a suite of utility scripts designed to annotate speech datasets.
* [The Parler-TTS organization](https://huggingface.co/parler-tts) - where you can find the annotated datasets as well as the future checkpoints.

## Citation

If you found this repository useful, please consider citing this work and also the original Stability AI paper:

```
@misc{lacombe-etal-2024-parler-tts,
  author = {Yoach Lacombe and Vaibhav Srivastav and Sanchit Gandhi},
  title = {Parler-TTS},
  year = {2024},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/huggingface/parler-tts}}
}
```

```
@misc{lyth2024natural,
      title={Natural language guidance of high-fidelity text-to-speech with synthetic annotations},
      author={Dan Lyth and Simon King},
      year={2024},
      eprint={2402.01912},
      archivePrefix={arXiv},
      primaryClass={cs.SD}
}
```

## License

This model is permissively licensed under the Apache 2.0 license.