ByteDance
Posted 2mo ago

豆包语音大模型算法工程师 - Data语音

ByteDance
Beijing, Beijing, China
OnsiteFull Time
Responsibilities
  • researching tts
  • optimizing models
  • deploying services
Requirements
  • Develop and optimize TTS and audio-generation large-model algorithms
  • Implement model lightweighting and inference acceleration, and drive deployment in production scenarios
Technical tools mentioned
PyTorchC++CUDATensorRTONNXVALL-ESeedTTSCosyVoiceDiTAR

Job description

团队介绍:Data-语音团队致力于语音、音乐等大模型AIGC技术的研发和产品创新,我们的使命是通过多模态AIGC音频技术赋能内容创作和语音交互,让内容生产、消费与互动变得更加简单、沉浸、多元化。当前团队已有成熟的大模型语音合成、语音理解、音乐生成等多模态音频技术,并持续迭代大模型端到端语音交互方案等,一方面通过中台形式服务于公司众多业务线,例如豆包/Dola、抖音/TikTok、剪映/CapCut、番茄小说、海绵音乐等产品;另一方面,团队已经通过字节跳动旗下的火山引擎开放平台-语音技术和火山方舟平台,向众多企业开放技术成熟稳定的能力和服务。

1、负责下一代语音合成(TTS)、音频生成大模型(Seed Audio)与语音大模型核心技术的研发,包括但不限于Zero-Shot TTS、声音克隆(Voice Cloning)、跨语种合成及高表现力语音生成;
2、持续优化合成语音的音质、自然度、情感表现力,并针对端/云侧业务需求,进行模型轻量化、稳定性调优与推理加速;
3、主导音频生成技术在字节跳动业务场景的落地与交付,跟进并解决真实复杂场景下的业务诉求,包括豆包、抖音、剪映等;
4、进行音频生成技术的Model-as-a-service应用探索,通过火山方舟进行国内和国外重点场景的业务应用能力建设,让语音技术从研究走向应用。

1、计算机科学、信号处理、人工智能等相关专业;有扎实的数学基础和算法/数据结构功底;
2、深耕语音合成领域,熟悉并掌握主流的TTS算法架构,熟悉基于大模型的音频生成技术栈(如VALL-E,SeedTTS,CosyVoice,DiTAR等),有离散音频表征(Audio Codec)、自回归/非自回归、扩散模型等研发经验者优先;
3、具备出色的工程能力和落地意识,熟练掌握PyTorch等深度学习框架;熟悉C++/CUDA或有TensorRT/ONNX推理优化、模型量化裁剪经验者优先;
4、加分项:在Interspeech、ICASSP、NeurIPS、ICML、ACL等会议上发表过语音/音频生成相关论文,或在相关技术竞赛中取得优异名次;
5、具备强烈的责任心与自驱力,能敏锐洞察业务需求,将技术突破转化为实际的产品价值。

About ByteDance

Developing AI-driven content platforms and mobile applications.

Similar jobs

Speech Algorithm Engineer roles near Beijing, Beijing
6d
Save
Mark Applied
Hide
语音算法工程师
Beijing, Beijing, China
OnsiteFull Time
Duxiaoman
Duxiaoman: A financial technology providing consumer credit and wealth solutions.
Experience in speech or multimodal large models; understanding of ASR, TTS, VAD, dialogue systems, data processing, and model architectures; Python, Linux, and deep learning framework proficiency.
Python, Linux, ASR, TTS, VAD, Fun-Audio-Chat, Moshi, Qwen3Omni
2w
Save
Mark Applied
Hide
混元大模型语音算法工程师(北京/上海)
Beijing or Shanghai or Shenzhen
OnsiteFull Time
Tencent
TencentHKEX: 0700: Provides integrated internet services, digital entertainment, and cloud technology.
4+ YOEExperience in speech/audio large-model development, strong coding and algorithm skills, proficiency in Python/C/C++, familiarity with model training frameworks, and solid math/signal processing background.
Python, C, C++, PyTorch, Megatron, DeepSpeed
3w
Save
Mark Applied
Hide
语音算法岗位-2027届
Beijing or Chongqing or Chengdu
OnsiteFull Time
Mashang Consumer Finance
Mashang Consumer Finance: A licensed financial institution providing digital consumer lending services.
Master's degree in computer science/electronic information/artificial intelligence, speech algorithm project experience, proficient in Python and basic AI coding, publications a plus; targeted to 2027 graduates.
Python
1mo
Save
Mark Applied
Hide
语音算法工程师实习生
Shenzhen or Beijing
OnsiteInternship
SenseTime
SenseTimeHKEX: 0020: Develops artificial intelligence software and computer vision technology.
Graduate-level in AI/ML/signal processing/computer science, familiarity with speech models and toolkits, strong research and PyTorch implementation skills, proficiency in Python/C/C++/Java/Shell, and experience with speech synthesis/recognition techniques.
kaldi, K2, wenet, espnet, whisper, FunASR, PyTorch, Python, C, C++, Java, Shell, RNN-T, conformer, CTC, SSL, LLM, diffusion, SpearTTS, ChatTTS, IndexTTS, CosyVoice
1mo
Save
Mark Applied
Hide
语音算法工程师-语音感知-北京
Beijing, Beijing, China
OnsiteFull Time
Li Auto
Li AutoNASDAQ / HKEX: LI / 2015: Designing and manufacturing premium smart electric vehicles.
Master's degree or above in AI/electronic information/computer science; expertise in wake-word, speech recognition, multimodal large models; proficiency in C/C++ and Python; research publications preferred.
C, C++, Python
3mo
Save
Mark Applied
Hide
语音算法工程师
Shenzhen or Beijing
OnsiteFull Time
SenseTime
SenseTimeHong Kong Stock Exchange: 0020: AI software provider of computer vision and deep learning.
Graduate degree in AI/ML/signal processing/computer science, strong ML and signal processing foundation, experience with speech recognition/synthesis models and toolkits, proficiency in PyTorch and programming in Python/C/C++/Java, publication or competition track record preferred.
RNN-T, conformer, CTC, Kaldi, K2, WeNet, ESPnet, Whisper, FunASR, PyTorch, SSL, LLM, diffusion, Python, C, C++, Java, Shell, SpearTTS, ChatTTS, IndexTTS, CosyVoice
3mo
Save
Mark Applied
Hide
全双工语音算法专家
Beijing, Beijing, China
OnsiteFull Time
ModelBest
ModelBest: Develops efficient, on-device large language models and edge AI.
3+ YOEMaster's in computer science or related,3+ years voice algorithm development,experience with ASR/TTS/LLM/end-to-end speech models,proficiency with TensorFlow or PyTorch and Python,strong engineering and communication skills.
TensorFlow, PyTorch, Python
3mo
Save
Mark Applied
Hide
实习-语音大模型算法工程师(TTS)
Beijing or Shanghai
OnsiteInternship
Nio
NioNYSE: NIO: Designs and manufactures smart premium electric vehicles.
Conduct TTS research and engineering including data processing, pretraining, fine-tuning, RL, and model evaluation; bachelor\u0002s degree in relevant fields required; strong communication and learning ability.