MOSS-VoiceGenerator

MOSS-TTS Family

Overview

MOSS‑TTS Family is an open‑source speech and sound generation model family from MOSI.AI and the OpenMOSS team. 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.

Introduction

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.

• 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.
• 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.
• 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.
• 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.
• 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.

Released Models

MOSS Voice Generator Model Card

MOSS Voice Generator is an open-source voice generation system designed to enable the creation of custom speaker timbres from free-form textual descriptions. This model allows users to generate voices that reflect specific characters, personalities, and emotions. It is particularly notable for its ability to produce speech with natural-sounding emotional expressiveness, providing a realistic and nuanced listening experience. As an open-source tool, MOSS Voice Generator is suitable for a variety of applications, such as audiobooks, game dubbing, role-playing agents, and conversational assistants.

1. Overview

1.1 TTS Family Positioning

MOSS Voice Generator is a high-fidelity voice design tool within the broader TTS Family. It specializes in crafting expressive and natural-sounding voices from textual descriptions. Unlike traditional TTS systems relying on predefined voices or reference audio, MOSS Voice Generator enables zero-shot voice design, allowing for the creation of customized voices for a variety of applications, such as characters, audiobooks, games, or virtual assistants. Additionally, it can serve as a voice design layer for other TTS systems, addressing the challenge of finding suitable reference audio and improving integration and performance.

Key Capabilities * Highly expressive emotional delivery: Aimed at generating voices with dynamic and nuanced emotional performances, allowing for natural shifts in tone, pace, and emotion. * Human-Like Naturalness : Indistinguishable from real human speech with authentic breathing, pauses, and vocal nuances * Multilingual Support : High-quality synthesis in Chinese and English

1.2 Model Architecture

MOSS Voice Generator employs MossTTSDelay architecture (see moss_tts_delay/README.md for more details), where voice description instructions and the text to be synthesized are concatenated and jointly tokenized as input to drive speech generation, enabling unified modeling of timbre design, style control, and content synthesis. Through instruction-timbre alignment, the model learns the correspondence between textual descriptions and acoustic features, allowing it to generate high-fidelity speech with target timbre, emotion, and style directly from free-form text prompts—without requiring any reference audio.

1.3 Released Model

Recommended decoding hyperparameters | Model | audio_temperature | audio_top_p | audio_top_k | audio_repetition_penalty | |---|---:|---:|---:|---:| | MOSS-VoiceGenerator | 1.5 | 0.6 | 50 | 1.1 |

2. Quick Start

Environment Setup

We recommend a clean, isolated Python environment with Transformers 5.0.0 to avoid dependency conflicts.

conda create -n moss-tts python=3.12 -y
conda activate moss-tts

Install all required dependencies:

git clone https://github.com/OpenMOSS/MOSS-TTS.git
cd MOSS-TTS
pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e .

(Optional) Install FlashAttention 2

For better speed and lower GPU memory usage, you can install FlashAttention 2 if your hardware supports it.

pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[flash-attn]"

If your machine has limited RAM and many CPU cores, you can cap build parallelism:

MAX_JOBS=4 pip install --extra-index-url https://download.pytorch.org/whl/cu128 -e ".[flash-attn]"

Notes: - Dependencies are managed in pyproject.toml, which currently pins torch==2.9.1+cu128 and torchaudio==2.9.1+cu128. - If FlashAttention 2 fails to build on your machine, you can skip it and use the default attention backend. - FlashAttention 2 is only available on supported GPUs and is typically used with torch.float16 or torch.bfloat16.

Basic Usage

import importlib.util
from pathlib import Path
import torch
import torchaudio
from transformers import AutoModel, AutoProcessor
# Disable the broken cuDNN SDPA backend
torch.backends.cuda.enable_cudnn_sdp(False)
# Keep these enabled as fallbacks
torch.backends.cuda.enable_flash_sdp(True)
torch.backends.cuda.enable_mem_efficient_sdp(True)
torch.backends.cuda.enable_math_sdp(True)

pretrained_model_name_or_path = "OpenMOSS-Team/MOSS-VoiceGenerator"
device = "cuda" if torch.cuda.is_available() else "cpu"
dtype = torch.bfloat16 if device == "cuda" else torch.float32

def resolve_attn_implementation() -> str:
    # Prefer FlashAttention 2 when package + device conditions are met.
    if (
        device == "cuda"
        and importlib.util.find_spec("flash_attn") is not None
        and dtype in {torch.float16, torch.bfloat16}
    ):
        major, _ = torch.cuda.get_device_capability()
        if major >= 8:
            return "flash_attention_2"

    # CUDA fallback: use PyTorch SDPA kernels.
    if device == "cuda":
        return "sdpa"

    # CPU fallback.
    return "eager"


attn_implementation = resolve_attn_implementation()
print(f"[INFO] Using attn_implementation={attn_implementation}")

processor = AutoProcessor.from_pretrained(
    pretrained_model_name_or_path,
    trust_remote_code=True,
    normalize_inputs=True, # normalize text and instruction input
)
processor.audio_tokenizer = processor.audio_tokenizer.to(device)


# ====== Batch demo  ======
text1="哎呀，我的老腰啊，这年纪大了就是不行了。"
instruction1="疲惫沙哑的老年声音缓慢抱怨，带有轻微呻吟。"

text2="亲爱的观众们，今天我要为大家做一道传说中的龙须面，这道面条细如发丝，需要极其精湛的手艺才能制作成功，请大家仔细观看我的每一个动作。"
instruction2="热情的美食节目主持人，语调生动活泼，充满对美食的热爱和专业精神。"

text3="Hey there, stranger! What brings you to our humble town? Looking for a good drink or a tall tale?"
instruction3="Hearty, jovial tavern owner's voice, loud and welcoming with a slightly gruff, friendly tone in American English, radiating warmth and hospitality."

text4="The quick brown fox jumps over the lazy dog."
instruction4="Clear, neutral voice for phonetic practice, even tempo and precise articulation in standard American English, emphasizing clarity of each word."

conversations = [
    [processor.build_user_message(text=text1, instruction=instruction1)],
    [processor.build_user_message(text=text2, instruction=instruction2)],
    [processor.build_user_message(text=text3, instruction=instruction3)],
    [processor.build_user_message(text=text4, instruction=instruction4)],
]

model = AutoModel.from_pretrained(
    pretrained_model_name_or_path,
    trust_remote_code=True,
    attn_implementation=attn_implementation,
    torch_dtype=dtype,
).to(device)
model.eval()

batch_size = 1

save_dir = Path("inference_root")
save_dir.mkdir(exist_ok=True, parents=True)
sample_idx = 0
with torch.no_grad():
    for start in range(0, len(conversations), batch_size):
        batch_conversations = conversations[start : start + batch_size]
        batch = processor(batch_conversations, mode="generation")
        input_ids = batch["input_ids"].to(device)
        attention_mask = batch["attention_mask"].to(device)

        outputs = model.generate(
            input_ids=input_ids,
            attention_mask=attention_mask,
        )

        for message in processor.decode(outputs):
            audio = message.audio_codes_list[0]
            out_path = save_dir / f"sample{sample_idx}.wav"
            sample_idx += 1
            torchaudio.save(out_path, audio.unsqueeze(0), processor.model_config.sampling_rate)

Input Types

UserMessage

Generation Hyperparameters

Note: MOSS-Voice-Generator is sensitive to decoding hyperparameters. See Released Models for recommended defaults.

3. Performance

MOSS Voice Generator demonstrates significant advantages in subjective evaluation. Using 160 internal test samples covering diverse voice styles, we established three independent evaluation dimensions: (1) Overall Preference - Which voice would you choose? (2) Instruction Following - Which audio best follows the instructions (gender, age, tone, emotion, accent, speed)? (3) Naturalness - Which audio sounds most like real human speech? Results show that MOSS Voice Generator outperforms all TTS systems that support zero predefined voices and customizable preview text across these three dimensions.