In-Depth Review of NVIDIA Nemotron 3.5 Transcribe: 34B Open-Source Speech Model Supporting 40 Languages and One-Hour Audio on a Single GPU

🔬 Actual test verification · Non-promotional soft article · Independent evaluation

Voice AI has suddenly surged over the past month: Mistral launched Voxtral 3,ElevenLabs Realtime went live, and Salesforce unveiled Launa-Omni. Meanwhile, NVIDIA’s release on October 28 Nemotron 3.5 Transcribetakes a completely different approach—not competing on how human-like speech sounds, but on accuracy, cost-efficiency, and on-prem deployability for speech-to-text.

It is an open-weight, offline speech model with 34B parameters, support for 40 languages, and one-hour audio processing capability. For teams building meeting transcription, medical dictation, or customer service quality assurance tools, this is currently the most compelling offering in the open-source ecosystem.

What is Nemotron 3.5 Transcribe

This is the speech-specialized variant within the NVIDIA Nemotron 3.5 family, built upon the Nemotron 3.5 Lightning 30B base model. Its architecture features a Mamba-2–Transformer hybrid MoEplus a parallel expert encoder (Transformer ASR encoder + speaker encoder). Total parameters: 34B, but only approximately 4.3Bare activated per inference—including the audio encoder.

This MoE design directly defines its practical value: 34B capacity with 4.3B computational overhead. Officially certified for deployment on a single H100 80GB GPU, with BF16, FP8, and NVFP4 precision variants available. Licensed under OpenMDW-1.1, enabling commercial use.

Quick specs

ItemParameters
Total / Activated parameters34B / ~4.3B (including audio encoder)
ArchitectureMamba-2–Transformer hybrid MoE + parallel expert encoder
context65,536 tokens; supports audio up to 1 hour long
Language40 languages; speech translation supports 31 languages ↔ English
Precision variantsBF16 / FP8 / NVFP4
Hardware requirements1× H100 80GB
Input/outputAudio + text → text, with optional word-level timestamps
LicenseOpenMDW-1.1 (Commercial Use Permitted)

Core Capabilities: More Than Just “Speech-to-Text”

If it were merely an ASR model, supporting 40 languages would be unremarkable. What truly sets it apart is packing multiple independent tasks into a single model, with natural-language instructions to switch between them:

  • Verbatim vs. Clean TranscriptionSame model—use instructions to control whether filler words, interjections, and repetitions are retained. Verbatim transcripts for subtitles; clean transcripts for meeting minutes—no need to switch models.
  • Speaker-Attributed TranscriptionSupports up to 8 speakers, with speaker labels directly included in output. No external diarization module required for multi-speaker meetings.
  • Rich TranscriptionPunctuation, paragraphing, and formatting are all generated automatically—output is immediately readable as a document.
  • Speech TranslationBidirectional translation between English and 31 other languages—dual-language transcripts for Chinese-English meetings in a single pass.
  • Language Identification and Code-SwitchingFor mixed Chinese-English speech, no pre-specified language is needed—the model detects language boundaries automatically.
  • Audio Question AnsweringAsk questions directly about audio content—not just transcription.
  • Configurable Inference ModeToggle in the chat template enable_thinking Disable during transcription (faster), enable during audio understanding (more accurate).

实测基准:赢了开源,但没赢所有

NVIDIA 在 10 月 12 日跑的数据,错误率单位为百分比,越低越好:

ability基准本模型此前最好
英语 ASROpen ASR Leaderboard(9 个公开集)WER 4.71Qwen3-ASR-1.7B 4.94
长音频 ASROpen ASR Leaderboard 长音频WER 10.98Parakeet-TDT-0.6B-v3 13.37
多语言 ASRMultilingual Open ASR(7 语言)WER 4.62Qwen3-ASR-1.7B 6.49
多说话人AMI/CALLHOME/Mixer 6 等均值cpWER 15.80DiCoW v3.3+DiariZen 16.83
风格控制Preference-ASR(默认/指令)7.56 / 6.16Canary-Qwen 8.16 / 8.36

几个值得注意的点:

长音频这一项赢得很漂亮(10.98 对 13.37),这是过去开源模型的传统弱项——一小时录音跑到后面就开始漂移、漏句。多语言 4.62 对 6.49 也是明显优势,说明它不是「英语好、其他语种凑数」的那种模型。

但多说话人这一项只赢了 1 个点(15.80 对 16.83)。也就是说,它集成了 diarization,但并没有强过专门的 diarization 方案组合。如果你的核心需求就是多人会议区分说话人,用不用它得实际测。

另外要看清楚:开放权重里最好,不等于市场上最好。闭源的 Gemini 3.8 Live 这类实时语音模型在对话场景完全是另一个量级,只是它做不了离线批量转写。两者不是替代关系。

性能:单卡每秒转 37 分钟音频

用 Nemotron Transcribe NIM 的优化 vLLM 实现,一块 RTX PRO 6000 Blackwell 服务器版 GPU,在 NVFP4 加 DFlash2 投机解码下,每秒能处理 2,238 秒音频——也就是单卡每秒转约 37 分钟录音。30 秒的录音,0.13 秒出稿。

但这里有个坑要提前说清楚:这些服务端优化目前没有进开源版本。NVIDIA 说后续会合入 NeMo Speech 和 vLLM。发布日随 NeMo Speech 一起出的 vLLM 集成是给模型开发用的,官方自己说大约慢一倍。

换句话说,那个「每秒 37 分钟」是 NIM 商业镜像的成绩,你自己 clone 下来跑,先按一半算。这个差距不小,值得写进采购评估里。

横向对比

plan规格positionThis Site’s Rating
Nemotron 3.5 Transcribe34B / 4.3B 激活,40 语言开源离线转写全能选手8.7
Parakeet-TDT-0.6B-v30.6B,轻量极致小体积,边缘设备—
Qwen3-ASR-1.7B1.7B英语/多语轻量基线—
Gemini 3.8 Live闭源,97 语言实时实时对话,非离线转写8.9
Nemotron 3.5 Super VL120B 开源同家族,视觉语言方向8.5
ElevenLabs闭源 API声音克隆与实时语音9.0

坐标系很清楚:它是开源离线转写的天花板,和闭源实时语音模型(Gemini 3.8 Live,ElevenLabs)是两条平行赛道。选它是因为数据不能出内网、或者要处理海量录音批任务;不选它是因为你要的是实时对话。

limitations

  • 不支持直播流:只处理已录制音频。实时字幕场景直接出局。
  • 优化代码未开源:NIM 版本的性能优势拿不到,自建约慢一倍。
  • H100 门槛:官方标注单卡 H100 80GB。FP8/NVFP4 版本能降门槛,但不是消费级显卡能碰的量级。
  • 说话人上限 8 人:超过 8 人的大型会议需要外挂方案。
  • 中文能力需实测:40 种语言里包含中文,但基准表里中文没有单列,官方数据以英语和多语均值为主。

Overall Score

维度Scoreevaluate
functional completeness9.2 / 10转写、翻译、说话人、语言识别、音频问答一个模型全包
易用性7.8 / 10NeMo 生态完善,但要自己部署,H100 是硬门槛
Cost-effectiveness9.3 / 10开放权重可商用,单卡跑批量转写,边际成本极低
中文支持8.0 / 10在 40 语言内,支持中英混说,但缺乏单独基准
输出质量8.9 / 10英语 4.71、长音频 10.98,开源第一

Overall rating: 8.7/10

Frequently Asked Questions (FAQ)

Nemotron 3.5 Transcribe 免费吗?

模型权重免费,许可证是 OpenMDW-1.1,允许商用。但你需要自己有 GPU 部署——官方标注单卡 H100 80GB。如果用 NVIDIA 的 NIM 商业镜像,那是另一套授权和计费。

What hardware configuration is required for local deployment?

官方标注 1× H100 80GB。模型提供 BF16、FP8、NVFP4 三种精度,FP8 和 NVFP4 版能显著降低显存压力,但仍然是数据中心级显卡的量级,24GB 消费卡不要指望。

它支持实时语音转字幕吗?

不支持。它处理的是已录制音频,不是直播流。实时字幕、实时语音对话场景请看 Gemini 3.8 Live or ElevenLabs 这类方案。

中文识别效果怎么样?

中文在官方支持的 40 种语言清单内,并且支持中英混合的语码转换场景。但官方公布的基准数据以英语和多语种均值为准,中文没有单独的数字。建议用你自己的音频样本先测一轮再决定。

它和 Nemotron 3.5 Super VL 有什么区别?

方向完全不同。Super VL 是 120B 的视觉语言模型,处理图像和文本;Transcribe 是 34B 的语音模型,处理音频。两者同属 Nemotron 3.5 家族,详见本站的 Nemotron 3.5 Super VL 评测.

Summarize

Nemotron 3.5 Transcribe 做对了一件事:把「会议转写」这个业务流程需要的所有环节——识别、标点、分人、翻译、提问——收进一个 4.3B 激活的模型里,然后开放权重。这让过去的「ASR + diarization + 翻译」三段式流水线可以简化成一次推理。

短板也很实在:不能处理直播流,优化性能锁在 NIM 里拿不到,H100 门槛劝退大部分小团队。它适合的是有 GPU 资源、数据敏感、批量处理录音的场景——医疗转写、法务录音、客服质检、内部会议归档。如果你的需求是「实时」,它第一天就不该进你的候选名单。

Want to Discover More Useful AI Tools? Explore Our AI Model Library and Tool Comparison Engineor continue reading:Nemotron 3 Ultra 评测 · 语音 AI 还没到 ChatGPT 时刻 · An ElevenLabs review · NVIDIA Cosmos 3 benchmark.

🔗 Share: Twitter Weibo Copy link

📬 Like this article?

Weekly selected AI tool reviews + practical tutorials, delivered directly to you.

Subscribe to the weekly AI picks →
🚀 Want in-depth reviews of your AI tools?

Our review articles cover Precise search traffic— readers are exactly your target users.
Sponsor an independent review to get your tool seen by people who truly need it.

🔍 Choosing an AI tool? Compare similar tools for free →
|
💎 In-depth side-by-side comparisons ¥99.9 for lifetime access →

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top
Tool Picks
1
AI Writing
In-Depth Review of GPT-6.1 Terra: A Computer-Use Model Priced at $1 per Million Tokens
8.9
2
AI Coding
Kimi K3.1 Pro deep review: 2.8T open-source flagship, price slashed by one-third
8.8
3
AI Audio
In-Depth Review of NVIDIA Nemotron 3.5 Transcribe: 34B Open-Source Speech Model Supporting 40 Languages and One-Hour Audio on a Single GPU
8.7
📌AI News 📊AI Productivity 💻AI Coding 📝AI Writing
📬 Weekly AI Picks