Update README.md
Browse files
README.md
CHANGED
|
@@ -1,217 +1,219 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
<
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
<a href="https://
|
| 19 |
-
<a href="https://
|
| 20 |
-
|
| 21 |
-
<a href="https://
|
| 22 |
-
|
| 23 |
-
<a href="https://
|
| 24 |
-
<a href="https://
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
- **MOSS‑
|
| 42 |
-
- **MOSS‑
|
| 43 |
-
- **MOSS‑
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
|
| 51 |
-
|
| 52 |
-
| **MOSS
|
| 53 |
-
|
|
| 54 |
-
| **MOSS‑
|
| 55 |
-
| **MOSS‑
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
**
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
- **
|
| 81 |
-
- **
|
| 82 |
-
|
| 83 |
-
**
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
| 98 |
-
|
| 99 |
-
|
|
| 100 |
-
|
| 101 |
-
|
| 102 |
-
|
| 103 |
-
|
| 104 |
-
|
| 105 |
-
|
| 106 |
-
|
| 107 |
-
|
| 108 |
-
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
| 123 |
-
|
| 124 |
-
|
| 125 |
-
|
| 126 |
-
|
| 127 |
-
|
| 128 |
-
```
|
| 129 |
-
|
| 130 |
-
|
| 131 |
-
|
| 132 |
-
|
| 133 |
-
|
| 134 |
-
```
|
| 135 |
-
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
-
|
| 140 |
-
|
| 141 |
-
|
| 142 |
-
|
| 143 |
-
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
import
|
| 148 |
-
import
|
| 149 |
-
|
| 150 |
-
|
| 151 |
-
|
| 152 |
-
#
|
| 153 |
-
torch.backends.cuda.
|
| 154 |
-
|
| 155 |
-
torch.backends.cuda.
|
| 156 |
-
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
]
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
| 182 |
-
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
|
| 194 |
-
|
| 195 |
-
|
| 196 |
-
|
| 197 |
-
|
| 198 |
-
|
| 199 |
-
|
| 200 |
-
|
| 201 |
-
|
| 202 |
-
|
| 203 |
-
|
| 204 |
-
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
|
| 208 |
-
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
|
| 212 |
-
|
| 213 |
-
|
| 214 |
-
|
| 215 |
-
|
|
| 216 |
-
|
| 217 |
-
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- text-to-audio
|
| 5 |
+
---
|
| 6 |
+
# MOSS-TTS Family
|
| 7 |
+
|
| 8 |
+
<br>
|
| 9 |
+
|
| 10 |
+
<p align="center">
|
| 11 |
+
|
| 12 |
+
<img src="https://speech-demo.oss-cn-shanghai.aliyuncs.com/moss_tts_demo/tts_readme_imgaes_demo/openmoss_x_mosi" height="50" align="middle" />
|
| 13 |
+
</p>
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
<div align="center">
|
| 18 |
+
<a href="https://github.com/OpenMOSS/MOSS-TTS/tree/main"><img src="https://img.shields.io/badge/Project%20Page-GitHub-blue"></a>
|
| 19 |
+
<a href="https://modelscope.cn/collections/OpenMOSS-Team/MOSS-TTS"><img src="https://img.shields.io/badge/ModelScope-Models-lightgrey?logo=modelscope&"></a>
|
| 20 |
+
<a href="https://mosi.cn/#models"><img src="https://img.shields.io/badge/Blog-View-blue?logo=internet-explorer&"></a>
|
| 21 |
+
<a href="https://github.com/OpenMOSS/MOSS-TTS"><img src="https://img.shields.io/badge/Arxiv-Coming%20soon-red?logo=arxiv&"></a>
|
| 22 |
+
|
| 23 |
+
<a href="https://studio.mosi.cn"><img src="https://img.shields.io/badge/AIStudio-Try-green?logo=internet-explorer&"></a>
|
| 24 |
+
<a href="https://studio.mosi.cn/docs/moss-tts"><img src="https://img.shields.io/badge/API-Docs-00A3FF?logo=fastapi&"></a>
|
| 25 |
+
<a href="https://x.com/Open_MOSS"><img src="https://img.shields.io/badge/Twitter-Follow-black?logo=x&"></a>
|
| 26 |
+
<a href="https://discord.gg/fvm5TaWjU3"><img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&"></a>
|
| 27 |
+
</div>
|
| 28 |
+
|
| 29 |
+
## Overview
|
| 30 |
+
MOSS‑TTS Family is an open‑source **speech and sound generation model family** from [MOSI.AI](https://mosi.cn/#hero) and the [OpenMOSS team](https://www.open-moss.com/). It is designed for **high‑fidelity**, **high‑expressiveness**, and **complex real‑world scenarios**, covering stable long‑form speech, multi‑speaker dialogue, voice/character design, environmental sound effects, and real‑time streaming TTS.
|
| 31 |
+
|
| 32 |
+
|
| 33 |
+
## Introduction
|
| 34 |
+
|
| 35 |
+
<p align="center">
|
| 36 |
+
<img src="https://speech-demo.oss-cn-shanghai.aliyuncs.com/moss_tts_demo/tts_readme_imgaes_demo/moss_tts_family_arch.jpeg" width="85%" />
|
| 37 |
+
</p>
|
| 38 |
+
|
| 39 |
+
When a single piece of audio needs to **sound like a real person**, **pronounce every word accurately**, **switch speaking styles across content**, **remain stable over tens of minutes**, and **support dialogue, role‑play, and real‑time interaction**, a single TTS model is often not enough. The **MOSS‑TTS Family** breaks the workflow into five production‑ready models that can be used independently or composed into a complete pipeline.
|
| 40 |
+
|
| 41 |
+
- **MOSS‑TTS**: MOSS-TTS is the flagship production TTS foundation model, centered on high-fidelity zero-shot voice cloning with controllable long-form synthesis, pronunciation, and multilingual/code-switched speech. It serves as the core engine for scalable narration, dubbing, and voice-driven products.
|
| 42 |
+
- **MOSS‑TTSD**: MOSS-TTSD is a production long-form dialogue model for expressive multi-speaker conversational audio at scale. It supports long-duration continuity, turn-taking control, and zero-shot voice cloning from short references for podcasts, audiobooks, commentary, dubbing, and entertainment dialogue.
|
| 43 |
+
- **MOSS‑VoiceGenerator**: MOSS-VoiceGenerator is an open-source voice design model that creates speaker timbres directly from free-form text, without reference audio. It unifies timbre design, style control, and content synthesis, and can be used standalone or as a voice-design layer for downstream TTS.
|
| 44 |
+
- **MOSS‑SoundEffect**: MOSS-SoundEffect is a high-fidelity text-to-sound model with broad category coverage and controllable duration for real content production. It generates stable audio from prompts across ambience, urban scenes, creatures, human actions, and music-like clips for film, games, interactive media, and data synthesis.
|
| 45 |
+
- **MOSS‑TTS‑Realtime**: MOSS-TTS-Realtime is a context-aware, multi-turn streaming TTS model for real-time voice agents. By conditioning on dialogue history across both text and prior user acoustics, it delivers low-latency synthesis with coherent, consistent voice responses across turns.
|
| 46 |
+
|
| 47 |
+
|
| 48 |
+
## Released Models
|
| 49 |
+
|
| 50 |
+
| Model | Architecture | Size | Model Card | Hugging Face |
|
| 51 |
+
|---|---|---:|---|---|
|
| 52 |
+
| **MOSS-TTS** | MossTTSDelay | 8B | [moss_tts_model_card.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/docs/moss_tts_model_card.md) | 🤗 [Huggingface](https://huggingface.co/OpenMOSS-Team/MOSS-TTS) |
|
| 53 |
+
| | MossTTSLocal | 1.7B | [moss_tts_model_card.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/docs/moss_tts_model_card.md) | 🤗 [Huggingface](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-Local-Transformer) |
|
| 54 |
+
| **MOSS‑TTSD‑V1.0** | MossTTSDelay | 8B | [moss_ttsd_model_card.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/docs/moss_ttsd_model_card.md) | 🤗 [Huggingface](https://huggingface.co/OpenMOSS-Team/MOSS-TTSD-v1.0) |
|
| 55 |
+
| **MOSS‑VoiceGenerator** | MossTTSDelay | 1.7B | [moss_voice_generator_model_card.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/docs/moss_voice_generator_model_card.md) | 🤗 [Huggingface](https://huggingface.co/OpenMOSS-Team/MOSS-Voice-Generator) |
|
| 56 |
+
| **MOSS‑SoundEffect** | MossTTSDelay | 8B | [moss_sound_effect_model_card.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/docs/moss_sound_effect_model_card.md) | 🤗 [Huggingface](https://huggingface.co/OpenMOSS-Team/MOSS-SoundEffect) |
|
| 57 |
+
| **MOSS‑TTS‑Realtime** | MossTTSRealtime | 1.7B | [moss_tts_realtime_model_card.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/docs/moss_tts_realtime_model_card.md) | 🤗 [Huggingface](https://huggingface.co/OpenMOSS-Team/MOSS-TTS-Realtime) |
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
# MOSS-SoundEffect
|
| 61 |
+
|
| 62 |
+
**MOSS-SoundEffect** is the **environment sound & sound effect generation model** in the **MOSS‑TTS Family**. It generates ambient soundscapes and concrete sound effects directly from text descriptions, and is designed to complement speech content with immersive context in production workflows.
|
| 63 |
+
|
| 64 |
+
|
| 65 |
+
## 1. Overview
|
| 66 |
+
|
| 67 |
+
### 1.1 TTS Family Positioning
|
| 68 |
+
|
| 69 |
+
MOSS-SoundEffect is designed as an audio generation backbone for creating high-fidelity environmental and action sounds from text, serving both scalable content pipelines and a strong research baseline for controllable audio generation.
|
| 70 |
+
|
| 71 |
+
**Design goals**
|
| 72 |
+
* **Coverage & richness**: broad sound taxonomy with layered ambience and realistic texture
|
| 73 |
+
* **Composability**: easy integration into creative pipelines (games/film/tools) and synthetic data generation setups
|
| 74 |
+
|
| 75 |
+
|
| 76 |
+
### 1.2 Key Capabilities
|
| 77 |
+
MOSS‑SoundEffect focuses on **contextual audio completion** beyond speech, enabling creators and systems to enrich scenes with believable acoustic environments and action‑level cues.
|
| 78 |
+
|
| 79 |
+
**What it can generate**
|
| 80 |
+
- **Natural environments**: e.g., “fresh snow crunching under footsteps.”
|
| 81 |
+
- **Urban environments**: e.g., “a sports car roaring past on the highway.”
|
| 82 |
+
- **Animals & creatures**: e.g., “early morning park with birds chirping in a quiet atmosphere.”
|
| 83 |
+
- **Human actions**: e.g., “clear footsteps echoing on concrete at a steady rhythm.”
|
| 84 |
+
|
| 85 |
+
**Why it matters**
|
| 86 |
+
- Completes **scene immersion** for narrative content, film/TV, documentaries, games, and podcasts.
|
| 87 |
+
- Supports **voice agents** and interactive systems that need ambient context, not just speech.
|
| 88 |
+
- Acts as the **sound‑design layer** of the MOSS‑TTS Family’s end‑to‑end workflow.
|
| 89 |
+
|
| 90 |
+
|
| 91 |
+
|
| 92 |
+
### 1.3 Model Architecture
|
| 93 |
+
**MOSS-SoundEffect** employs the **MossTTSDelay** architecture (see [moss_tts_delay/README.md](https://github.com/OpenMOSS/MOSS-TTS/blob/main/moss_tts_delay/README.md)), reusing the same discrete token generation backbone for audio synthesis. A text prompt (optionally with simple control tags such as **duration**) is tokenized and fed into the Delay-pattern autoregressive model to predict **RVQ audio tokens** over time. The generated tokens are then decoded by the audio tokenizer/vocoder to produce high-fidelity sound effects, enabling consistent quality and controllable length across diverse SFX categories.
|
| 94 |
+
|
| 95 |
+
|
| 96 |
+
|
| 97 |
+
### 1.4 Released Models
|
| 98 |
+
**Recommended decoding hyperparameters**
|
| 99 |
+
| Model | audio_temperature | audio_top_p | audio_top_k | audio_repetition_penalty |
|
| 100 |
+
|---|---:|---:|---:|---:|
|
| 101 |
+
| **MOSS-SoundEffect** | 1.5 | 0.6 | 50 | 1.2 |
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
## 2. Quick Start
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
### Environment Setup
|
| 108 |
+
|
| 109 |
+
We recommend a clean, isolated Python environment with **Transformers 5.0.0** to avoid dependency conflicts.
|
| 110 |
+
|
| 111 |
+
```bash
|
| 112 |
+
conda create -n moss-tts python=3.12 -y
|
| 113 |
+
conda activate moss-tts
|
| 114 |
+
```
|
| 115 |
+
|
| 116 |
+
Install all required dependencies:
|
| 117 |
+
|
| 118 |
+
```bash
|
| 119 |
+
git clone https://github.com/OpenMOSS/MOSS-TTS.git
|
| 120 |
+
cd MOSS-TTS
|
| 121 |
+
pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e .
|
| 122 |
+
```
|
| 123 |
+
|
| 124 |
+
#### (Optional) Install FlashAttention 2
|
| 125 |
+
|
| 126 |
+
For better speed and lower GPU memory usage, you can install FlashAttention 2 if your hardware supports it.
|
| 127 |
+
|
| 128 |
+
```bash
|
| 129 |
+
pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[flash-attn]"
|
| 130 |
+
```
|
| 131 |
+
|
| 132 |
+
If your machine has limited RAM and many CPU cores, you can cap build parallelism:
|
| 133 |
+
|
| 134 |
+
```bash
|
| 135 |
+
MAX_JOBS=4 pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[flash-attn]"
|
| 136 |
+
```
|
| 137 |
+
|
| 138 |
+
Notes:
|
| 139 |
+
- Dependencies are managed in `pyproject.toml`, which currently pins `torch==2.9.1+cu128` and `torchaudio==2.9.1+cu128`.
|
| 140 |
+
- If FlashAttention 2 fails to build on your machine, you can skip it and use the default attention backend.
|
| 141 |
+
- FlashAttention 2 is only available on supported GPUs and is typically used with `torch.float16` or `torch.bfloat16`.
|
| 142 |
+
|
| 143 |
+
|
| 144 |
+
### Basic Usage
|
| 145 |
+
|
| 146 |
+
```python
|
| 147 |
+
import os
|
| 148 |
+
from pathlib import Path
|
| 149 |
+
import torch
|
| 150 |
+
import torchaudio
|
| 151 |
+
from transformers import AutoModel, AutoProcessor
|
| 152 |
+
# Disable the broken cuDNN SDPA backend
|
| 153 |
+
torch.backends.cuda.enable_cudnn_sdp(False)
|
| 154 |
+
# Keep these enabled as fallbacks
|
| 155 |
+
torch.backends.cuda.enable_flash_sdp(True)
|
| 156 |
+
torch.backends.cuda.enable_mem_efficient_sdp(True)
|
| 157 |
+
torch.backends.cuda.enable_math_sdp(True)
|
| 158 |
+
|
| 159 |
+
|
| 160 |
+
pretrained_model_name_or_path = "OpenMOSS-Team/MOSS-SoundEffect"
|
| 161 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 162 |
+
dtype = torch.bfloat16 if device == "cuda" else torch.float32
|
| 163 |
+
|
| 164 |
+
processor = AutoProcessor.from_pretrained(
|
| 165 |
+
pretrained_model_name_or_path,
|
| 166 |
+
trust_remote_code=True,
|
| 167 |
+
)
|
| 168 |
+
processor.audio_tokenizer = processor.audio_tokenizer.to(device)
|
| 169 |
+
|
| 170 |
+
text_1 = "雷声隆隆，雨声淅沥。"
|
| 171 |
+
text_2 = "清晰脚步声在水泥地面回响，节奏稳定。"
|
| 172 |
+
|
| 173 |
+
conversations = [
|
| 174 |
+
[processor.build_user_message(ambient_sound=text_1)],
|
| 175 |
+
[processor.build_user_message(ambient_sound=text_2)]
|
| 176 |
+
]
|
| 177 |
+
|
| 178 |
+
model = AutoModel.from_pretrained(
|
| 179 |
+
pretrained_model_name_or_path,
|
| 180 |
+
trust_remote_code=True,
|
| 181 |
+
attn_implementation="sdpa",
|
| 182 |
+
torch_dtype=dtype,
|
| 183 |
+
).to(device)
|
| 184 |
+
model.eval()
|
| 185 |
+
|
| 186 |
+
batch_size = 1
|
| 187 |
+
|
| 188 |
+
messages = []
|
| 189 |
+
save_dir = Path("inference_root")
|
| 190 |
+
save_dir.mkdir(exist_ok=True, parents=True)
|
| 191 |
+
sample_idx = 0
|
| 192 |
+
with torch.no_grad():
|
| 193 |
+
for start in range(0, len(conversations), batch_size):
|
| 194 |
+
batch_conversations = conversations[start : start + batch_size]
|
| 195 |
+
batch = processor(batch_conversations, mode="generation")
|
| 196 |
+
input_ids = batch["input_ids"].to(device)
|
| 197 |
+
attention_mask = batch["attention_mask"].to(device)
|
| 198 |
+
|
| 199 |
+
outputs = model.generate(
|
| 200 |
+
input_ids=input_ids,
|
| 201 |
+
attention_mask=attention_mask,
|
| 202 |
+
max_new_tokens=4096,
|
| 203 |
+
)
|
| 204 |
+
|
| 205 |
+
for message in processor.decode(outputs):
|
| 206 |
+
audio = message.audio_codes_list[0]
|
| 207 |
+
out_path = save_dir / f"sample{sample_idx}.wav"
|
| 208 |
+
sample_idx += 1
|
| 209 |
+
torchaudio.save(out_path, audio.unsqueeze(0), processor.model_config.sampling_rate)
|
| 210 |
+
```
|
| 211 |
+
|
| 212 |
+
### Input Types
|
| 213 |
+
|
| 214 |
+
**UserMessage**
|
| 215 |
+
| Field | Type | Required | Description |
|
| 216 |
+
|---|---|---:|---|
|
| 217 |
+
| `ambient_sound` | `str` | Yes | Description of environment sound & sound effect |
|
| 218 |
+
| `tokens` | `int` | No | Expected number of audio tokens. **1s ≈ 12.5 tokens**. |
|
| 219 |
+
|