Granite Speech 5.0：更准确、更快速地转录、翻译和说话人归属

我们很高兴与大家分享 IBM Granite Speech 系列的三项更新：一款新的语音语言模型 granite-speech-5.0-2b，以及社区已在使用的两款模型的修订版本 granite-speech-5.0-470m-turboctc 和 granite-speech-4.1-2b-plus。它们共同覆盖了企业语音任务的全部需求：高精度转录与翻译、边缘端的超快速转录，以及在这些录音中，说话者是谁、何时发言与说了什么同样重要的场景。We're excited to share three updates to IBM's Granite Speech collection: a new speech-language model, granite-speech-5.0-2b, and revised versions of two models the community already uses, granite-speech-5.0-470m-turboctc and granite-speech-4.1-2b-plus. Together they cover the full range of enterprise speech work: transcription and translation with high accuracy, extremely fast transcription at the edge, and recordings where who spoke, and when, matter as much as what was said.

Granite Speech 5.0 2B：更准确、更稳健、控制力更强 Granite Speech 5.0 2B: more accurate, more robust, more control

granite-speech-5.0-2b 是 granite-speech-4.1-2b 的后继版本，参数量相同。它可转写英语、法语、德语、西班牙语、葡萄牙语和日语，并支持这些语言与英语之间的互译，还支持将英语翻译成意大利语和普通话。它使用 39 万小时的公开音频进行训练，此外还使用了为日语 ASR、关键词偏置和翻译构建的合成数据。granite-speech-5.0-2b is the successor to granite-speech-4.1-2b at the same parameter count. It transcribes English, French, German, Spanish, Portuguese, and Japanese, and translates to and from English in those languages, plus English to Italian and Mandarin. It was trained on 390,000 hours of public audio, plus synthetic data built for Japanese ASR, keyword biasing, and translation.

与 4.1 2B Speech 相比有哪些新内容：What's new compared with 4.1 2B Speech:

• 更准确、更稳健的英语 ASR：在英语短语音基准测试中准确率更高，并且能更好地处理噪声、混响和远距离麦克风语音。
  More accurate, more robust English ASR: Higher accuracy on English short-form benchmarks, and better handling of noise, reverberation, and distant-microphone speech.
• 语言 ID 提示：在提示中指定输入语言，以引导多语言转录。
  Language ID prompting: Name the input language in the prompt to steer multilingual transcription.
• 更适合长篇内容和对话场景的翻译：翻译在较长文本中也能保持准确，表达更加自然流畅，不再局限于简短的书面句子。
  Better long-form and conversational translation: Translation holds up for longer with more natural speech, not just short read sentences.
• 关键词列表最多可包含 2,000 个词条：在所有支持的语言中，转录和翻译时优先采用产品名称、人名和首字母缩略词。
  Keyword lists up to 2,000 terms: Bias transcription and translation toward product names, people, and acronyms, in every supported language.
• 按需添加标点和大小写：在提示中要求格式化输出；在我们的测试集上，大小写 F1 分数介于 89.7 至 99.5 之间。
  Punctuation and capitalization on request: Ask for formatted output in the prompt; capitalization F1 ranges from 89.7 to 99.5 across our test sets.

该模型基于 Apache 2.0 许可证发布，并支持在 Transformers、vLLM、llama.cpp 和适用于 Apple Silicon 的 mlx-audio 中使用。The model ships under Apache 2.0 with support in Transformers, vLLM, llama.cpp, and mlx-audio for Apple Silicon.

图表 1：granite-speech-5.0-2b 在顶尖开放权重 ASR 模型中，按英语平均 WER（5.06%）排名第 2。数值越低越好。结果为非官方数据，截至 2026 年 10 月 15 日，取自 Open ASR Leaderboard 的公开英语短音频测试集的平均值。Chart 1: granite-speech-5.0-2b ranks #2 among top open-weight ASR models on average English WER (5.06%). Lower is better. Results are unofficial, as of 15 October 2026, and averaged over the Open ASR Leaderboard’s public English short-form test sets.

Granite Speech 5.0 Turbo CTC：现已支持多语言，更加稳健，速度更快 Granite Speech 5.0 Turbo CTC: now multilingual, more robust, and even faster

我们在 8 月推出 Turbo CTC 时，它是一款仅支持英语、拥有 4.7 亿个参数、专为速度打造的编码器。本次修订保留了这一规模，并加入了用户最希望看到的三项功能：When we introduced Turbo CTC in August, it was an English-only, 470M-parameter encoder built for speed. This revision keeps that footprint and adds the three things users asked for most:

• 多语言转录：英语、日语、西班牙语、德语、法语、意大利语和葡萄牙语。
  Multilingual transcription: English, Japanese, Spanish, German, French, Italian, and Portuguese.
• 原生标点和大小写还原：转录内容生成后即可直接阅读，无需单独进行标点添加或大小写还原步骤。
  Native punctuation and capitalization: Transcripts come out ready to read, with no separate punctuation or truecasing step.
• 在复杂声学环境中更加稳健：针对混响音频进行专门训练，使 Turbo CTC 在回声较强的房间和远场录音中表现得更加稳健。
  More robust in difficult acoustics: Training that explicitly targets reverberant audio makes Turbo CTC more robust in echoey rooms and far-field recordings.

速度也更快。通过编译单个 Conformer 模块，并在整个网络堆栈中复用其计算图，我们缩短了预热时间，使 Turbo CTC 在 Open ASR Leaderboard 上的速度从发布时约 12,600 RTFx 提升至 19,870 RTFx：在准确率相同的情况下，速度约提升了 1.6 倍。It's also faster. By compiling a single Conformer block and reusing its graph across the stack, we cut warmup time and pushed Turbo CTC to 19,870 RTFx on the Open ASR Leaderboard, up from roughly 12,600 RTFx at launch: about 1.6x faster, at the same accuracy level.

图 2：Open ASR Leaderboard 上平均 WER 与 RTFx 的关系（英语短音频，9 个公开测试集，截至 2026 年 10 月）。WER 越低、RTFx 越高越好；RTFx 使用对数刻度。Chart 2: Average WER vs. RTFx on the Open ASR Leaderboard (English short-form, 9 public test sets, as of October 2026). Lower WER and higher RTFx are better; RTFx is on a log scale.

Granite Speech 4.1 2B Plus：更准确地识别谁说了什么，以及何时说的 Granite Speech 4.1 2B Plus: better at who said what, and when

更新后的 granite-speech-4.1-2b-plus 针对企业实际拥有的录音进行了调优：长时间、多位发言者参与的会议、客服通话和访谈。它保留了 4.1 这一名称，因为架构、提示格式和 API 均未改变，因此可直接替换当前模型，并采用了改进后的权重。此版本改进了四个方面：The updated granite-speech-4.1-2b-plus is tuned for the recordings enterprises actually have: meetings, support calls, and interviews with several speakers running long. It keeps the 4.1 name because the architecture, prompt format, and API are unchanged, so it's a drop-in replacement for the current model with improved weights. This revision improves four areas:

• 说话人归属：更准确地标注发言者。词级说话人区分错误率（WDER）下降 7%，带说话人归属的词错误率（cpWER）总体下降 7%。
  Speaker attribution: More accurate labeling of who is speaking. Word diarization error rate (WDER) drops 7%, and speaker-attributed WER (cpWER) drops 7% overall.
• 时间戳：平均时间戳偏移（AAS）减少了一半以上，使字幕、搜索和审核中的时间定位更加精准。
  Timestamps: Average timestamp shift (AAS) is cut by more than half, yielding tighter timing in captions, search, and review.
• 长语音性能：长篇测试集上的词错误率（WER）下降了 24%。
  Long-speech performance: WER on long-form test sets drops 24%.
• 关键词偏置：使用包含 20 个术语的关键词列表后，偏置关键词的错误率降低了 45%。
  Keyword biasing: Errors on biased keywords fall 45% with 20-term keyword lists.

表 1：所有指标均为错误率，因此越低越好。更多详细结果请参阅模型卡。Table 1: All metrics are error rates, so lower is better. For more detailed results, see the model card.

开始使用 Get started

不确定该选择哪个模型？从这里开始：Not sure which model to pick? Start here:

• 需要高精度转录、翻译或关键词偏置？请使用 granite-speech-5.0-2b。
  Need high transcription accuracy, translation, or keyword biasing? Use granite-speech-5.0-2b.
• 想要极致速度还是设备端转录？使用 granite-speech-5.0-470m-turboctc，还可以在浏览器中试用我们的 WebGPU 流式演示。
  Need raw speed or on-device transcription? Use granite-speech-5.0-470m-turboctc, and try it in the browser with our WebGPU streaming demo.
• 处理会议或多人通话？请使用 granite-speech-4.1-2b-plus。
  Working with meetings or multi-speaker calls? Use granite-speech-4.1-2b-plus.

这些模型均以 Apache 2.0 许可证发布，并原生支持 Transformers。每个模型卡都提供完整的评估结果、架构详情、训练数据和使用示例。对于需要额外风险检测的生产部署，我们建议搭配使用 Granite Guardian。These models are all released under an Apache 2.0 license with native support in Transformers. Full evaluation results, architecture details, training data, and usage examples are on each model card. We recommend pairing them with Granite Guardian for production deployments that need additional risk detection.

试试看，并告诉我们你的想法！Give them a try and let us know what you think!