distil-whisper/spgispeech-timestamped

---
license: other
task_categories:
- automatic-speech-recognition
language:
- en
extra_gated_prompt: |-
Your access to and use of the information in the Kensho Transcript Dataset (the “Content”), which is provided by Kensho Technologies, LLC, a subsidiary of S&P Global, Inc., (“Kensho”), shall be governed by the following terms and conditions of usage (“Terms of Usage”). The Content may be accessed only by persons who have been authorized to use this Content pursuant to their acceptance and acknowledgement of these Terms of Usage (in each case, an “Authorized User”). By providing your electronic signature at the end of these Terms of Usage, you represent that you are an Authorized User and that you accept these Terms of Usage and agree to be bound by them.
If you do not wish to be bound by these Terms of Usage, you must not use this Content. PLEASE READ THESE TERMS OF USAGE CAREFULLY BEFORE USING THIS CONTENT.
Section 1 – THE CONTENT
1.1 The Content is provided for academic research purposes and internal use only and must not be used to: assemble or create a database; construct or facilitate the construction of products which compete with the Content; identify or attempt to identify or contact any individual; or link to another dataset.
The Content, which is comprised of public earnings calls in audio and corresponding text format, and all accompanying derived products is proprietary to Kensho and its third-party content providers. You shall not modify the Content; create derivative works based on the Content, rewrite or reprocess the Content except as expressly provided herein. You must not publish, display, transfer or redistribute the Content or any portions or derivative versions thereof to anyone without prior written consent from Kensho. You agree not to contact Kensho or its affiliates concerning individuals whose information may be included in the Content.
1.2 Disclaimer. Content to which you are provided access, either directly or indirectly, from or on this Content will not have been reviewed or monitored by Kensho, and Kensho cannot and does not guarantee or make any representation or warranty, either express or implied, as to the accuracy, validity, timeliness, completeness or continued availability of any such content.
The Content is provided for your convenience only and is not a republication or reconfirmation of the opinion or information contained therein. The provision of the Content is without any obligation on the part of Kensho or its third-party content providers to review such or any liability or responsibility arising out of your use thereof. Kensho does not guarantee or make any representation or warranty, either express or implied, as to the accuracy, validity, timeliness, completeness or continued availability of any Content and shall not be liable for any errors, delays, or actions taken in reliance on information. In addition, the Content speaks only as of the date issued and is based on conference calls that may contain projections of other forward-looking statements. You should not rely on the Content as expressing Kensho’s opinion or as representing current information. None of Kensho or the third-party content providers has undertaken, and do not undertake any duty to update any Content or otherwise advise you of any changes in the Content.
1.3 Ownership of Third-Party Content. You acknowledge that all proprietary rights in the Content that are owned by Kensho or third party content providers shall remain the property of Kensho or such third party content providers, and you shall have no right or interest in such third party content except the rights to use such third party content in accordance with these Terms of Usage. Any additional rights not granted herein shall require a separate, direct agreement with Kensho. You acknowledge that the Content and third party content as compiled, prepared, selected and arranged by Kensho or its third party content providers constitutes an expenditure of substantial time, effort and money by Kensho and its third party content providers and constitutes valuable commercial property and/or trade secrets of Kensho and such third party content providers. Kensho retains all rights and remedies afforded under the copyright, trademark, service mark, patent and other laws of the United States and the States thereof, including without limitation any laws designed to protect proprietary or confidential information. You agree that you will not remove or modify any copyright notice, disclosures, disclaimers or other notification or trade name or marks of Kensho or the third party content providers that may appear in the Content or third party content and that any permitted reproduction and/or distribution of the Content or third party content shall contain such notices and/or marks as they appear in the Content or third party content. You may not use Kensho’s or the third-party content providers’ name or trademarks without the prior written consent of Kensho or such third-party content providers. Apart from the rights granted hereunder, no conveyance of ownership, right, title or interest is intended herein. Any additional rights require a separate agreement with Kensho.
1.4 Posted Guidelines. In addition to these Terms of Usage, when using this Content, you shall be subject to and agree to follow any posted notice, guidelines or rules, which may be posted and amended from time to time. Nothing on this Content shall be considered a recommendation or solicitation to buy or an offer to sell a security to any person in any jurisdiction.
1.5 Registration Data. In consideration of your use of this Content, you and/or your employer agree to: (a) provide true, accurate, current and complete Registration Data (as defined below in Section 3.1) to Kensho as prompted by the registration form completed prior to accessing the Content and (b) maintain and promptly update the Registration Data and to keep the same true, accurate, current and complete.
1.6 Right to Terminate User Access. Kensho reserves the right to limit, restrict and immediately terminate your access to and use of this Content at any time, in whole or in part, in its sole discretion and without notice.
Section 2 - DISCLAIMER OF WARRANTY AND LIMITATION OF LIABILITY
2.1 THE CONTENT IS PROVIDED “AS IS” AND “AS AVAILABLE” WITHOUT REPRESENTATION OR WARRANTY OF ANY KIND. USE OF THE CONTENT IS AT THE USER’S OWN RISK. IN NO EVENT SHALL KENSHO OR ITS THIRD-PARTY CONTENT PROVIDERS BE LIABLE FOR ANY DECISION MADE OR ACTION OR INACTION TAKEN IN RELIANCE ON ANY CONTENT, INCLUDING THIRD-PARTY CONTENT, INCLUDING YOUR HANDLING AND STORING OF THE CONTENT. KENSHO FURTHER EXPLICITLY DISCLAIMS, ANY WARRANTY OF ANY KIND, WHETHER EXPRESS OR IMPLIED, INCLUDING WARRANTIES OF ORIGINALITY, ACCURACY, COMPLETENESS, TIMELINESS, MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. KENSHO EXPRESSLY DISCLAIMS, AND YOU WAIVE, ANY LIABILITY THAT MAY ARISE FROM YOUR PUBLICATION OR PROVISION OF THE CONTENT TO A THIRD PARTY, OR ANY REPRESENTATION OR WARRANTY MADE BY YOU TO ANY THIRD PARTY, WHETHER OR NOT RELATED TO THE CONTENT. KENSHO, SUPPLIERS OF THIRD-PARTY CONTENT AND ANY OTHER THIRD PARTY WORKING WITH KENSHO SHALL NOT BE RESPONSIBLE OR LIABLE, DIRECTLY OR INDIRECTLY, FOR ANY DAMAGES OR LOSS (INCLUDING DIRECT, INDIRECT, INCIDENTAL, CONSEQUENTIAL AND ANY AND ALL OTHER FORMS OF DAMAGES OR LOSSES REGARDLESS OF THE FORM OF THE ACTION OR THE BASIS OF THE CLAIM) CAUSED OR ALLEGED TO BE CAUSED IN CONNECTION WITH YOUR USE OF THE CONTENT WHETHER OR NOT FORESEEABLE, EVEN IF KENSHO OR ANY OF THE SUPPLIERS OF THIRD-PARTY CONTENT OR OTHER THIRD PARTIES WORKING WITH KENSHO IN CONNECTION WITH THE CONTENT HAS BEEN ADVISED OF THE POSSIBILITY OR LIKELIHOOD OF SUCH DAMAGES.
2.2 THE CONTENT IS NOT INTENDED TO PROVIDE TAX, LEGAL, INSURANCE OR INVESTMENT ADVICE, AND NOTHING IN THE CONTENT SHOULD BE CONSTRUED AS AN OFFER TO SELL, A SOLICITATION OF AN OFFER TO BUY, OR A RECOMMENDATION FOR ANY SECURITY BY KENSHO OR ANY THIRD PARTY.
2.3 For third party demands, claims, actions, proceedings and liability for losses, damages, reasonable legal costs and other reasonable expenses of any nature, you agree to defend, indemnify and hold Kensho and its affiliates harmless, including its respective directors, officers, employees and agents from and against all claims to the extent arising from your access to and/or use of the Content, any failure by you to abide by the Terms of Usage, or breach of applicable law.
Section 3 - PRIVACY
3.1 Access and Collection. In order to access this Content, during the registration process, either you or your employer will be required to provide Kensho with certain information; including your name, employer or academic institution, and e-mail address (“Registration Data”). In addition, when you request or view Content, Kensho may obtain user identifiable information related to your request of, or access to, such Content (“Access Data”). For example, while you are accessing this Content, our Web servers may recognize your: (a) domain name; (b) ISP’s domain name; (c) IP address; (d) browser type; and (e) operating system. If you contact us with a technical question, we may collect certain information about your systems, including: (a) your browser type, version and settings (e.g., Java and cookie settings); (b) connectivity information (e.g., SSL/HTTPS compatibility, bandwidth capacity); and browser plug-in information (e.g., do you have Adobe, what is your media player, can you open Flash files, etc.).
3.2 Use of Your Information. Registration Data and Access Data may be used by Kensho for research and development purposes and to communicate with users and to troubleshoot any technical issues pertaining to the Content. You acknowledge that in the event that a separate agreement is required, Kensho may share Registration Data with its Affiliates (as defined below).
3.3 Disclosure of Your Information. Except as otherwise noted herein, Kensho will not disclose, rent or sell personal information collected from or about you without your permission. For the purposes specified in the preceding paragraph, we may transfer or disclose Registration Data and Access Data to S&P Global Inc. and its affiliates (“Kensho Affiliates”) and third parties who are contracted to perform services on behalf of Kensho, such as those who assist Kensho in bringing you this Content and providing you with certain features and functionality included within or accessible via this Content. We may also disclose Registration Data and Access Data to Kensho Affiliates and third parties in connection with their providing you access to this Content. Disclosures to these third parties will be subject to confidentiality agreements and, where required, governed by contract. Kensho may also be required to disclose information to governmental, regulatory or self-regulatory entities or agencies in response to regulatory inquiries or to comply with applicable laws, rules, regulations, orders, subpoenas or other legal processes.
3.4 Consent. By (a) agreeing to these Terms of Usage, or (b) by using this Content, and, in either case, providing any information that may be required, requested or otherwise collected by us as set forth above, you freely consent to Kensho processing your information in the United States and in other countries and territories for the purposes set out in these Terms of Usage, and you also consent to the transfer of your information for such purposes to any third party content provider wherever such entity may from time to time be located and to any third parties as described above and in accordance with applicable law and regulations. If you do not permit Kensho to collect any of your information or do not agree with any of the terms and conditions of these Terms of Usage, you should not use this Content and should exit this page and/or Content, as the case may be. If after registering with Kensho, you desire to withdraw the consent granted in this Section 3.4 for all future use of your information by Kensho, you must notify Kensho in writing at the address listed below in Section 3.8 and immediately cease use of this Content.
3.5 Inquiries. If you have any questions regarding these Terms of Usage or your information that is held by us, please contact Kensho in writing using the contact information provided below. If we receive a request regarding your personal information held by us, we will use reasonable means to provide you with such information that we can reasonably compile. You will be given the opportunity to rectify any inaccuracies in such information.
3.6 Encryption. Kensho may use encryption technology to protect certain transmissions of data to/from this Content, but e-mail and other communications, unless otherwise noted on this Content, are not encrypted to/from this Content. Therefore, you should not send any personal or identifying information, such as account numbers, credit card numbers, Social Security numbers, passwords, etc., to Kensho via e-mail. By utilizing e-mail or other electronic communication means you acknowledge that you have no expectation of privacy with respect to the information delivered thereby and that Kensho will not be responsible for any loss or damage that could result from interception by third parties of any information so sent.
3.7 Contact Information. In the event you have any questions regarding these Terms of Use, this Privacy Statement or to make any requests or queries regarding your information that is held by us you may contact us in writing at [e-mail] or Kensho Technologies LLC, Attn: General Counsel, 55 Water Street, New York, NY 10041.
Section 4 - MISCELLANEOUS
4.1 Entire Agreement. These Terms of Usage constitute the entire agreement of the parties hereto with respect to the subject matter hereof and supersede all prior agreements and undertakings, both written and oral, between the parties with respect to the subject matter hereof.
4.2 Severability. If any term or other provision of these Terms of Usage is invalid, illegal or incapable of being enforced by any law or public policy, all other terms and provisions of these Terms of Usage shall nevertheless remain in full force and effect so long as the economic or legal substance of the transactions contemplated hereby is not affected in any manner materially adverse to any party.
4.3 Governing Law; Forum. These Terms of Usage shall be governed in all respects by the laws of the State of New York, and any litigation arising out of or connected in any way with these Terms of Usage shall take place in a State or Federal court of competent jurisdiction in New York County, State of New York.
4.4 Waiver of Jury Trial. YOU WAIVE TO THE FULLEST EXTENT PERMITTED BY APPLICABLE LAW ANY RIGHT YOU MAY HAVE TO A TRIAL BY JURY WITH RESPECT TO ANY ACTIONS OR PROCEEDINGS DIRECTLY OR INDIRECTLY ARISING OUT OF, UNDER OR IN CONNECTION WITH THESE TERMS OF USAGE.
4.5 Conflict. In the event of a conflict between these Terms of Use and any other agreement with Kensho that relates to Third-Party Content, the more restrictive terms shall prevail.

extra_gated_fields:
Full name: text
Email: text
Institution: text
I accept the Terms of Usage: checkbox
---

# Distil Whisper: SPGISpeech With Timestamps

This is a variant of the [SPGISpeech](https://huggingface.co/datasets/kensho/spgispeech) dataset, augmented to return the pseudo-labelled Whisper
Transcriptions alongside the original dataset elements. The pseudo-labelled transcriptions were generated by
labelling the input audio data with the Whisper [large-v2](https://huggingface.co/openai/whisper-large-v2)
model with *greedy* sampling and timestamp prediction. For information on how the original dataset was curated, refer to the original
[dataset card](https://huggingface.co/datasets/kensho/spgispeech).

## Standalone Usage

First, install the latest version of the 🤗 Datasets package:

```bash
pip install --upgrade pip
pip install --upgrade datasets[audio]
```

The dataset can be downloaded and pre-processed on disk using the [`load_dataset`](https://huggingface.co/docs/datasets/v2.14.5/en/package_reference/loading_methods#datasets.load_dataset)
function:

```python
from datasets import load_dataset

dataset = load_dataset("distil-whisper/spgispeech", "L")
# take the first sample of the validation set
sample = dataset["validation"][0]
```

It can also be streamed directly from the Hub using Datasets' [streaming mode](https://huggingface.co/blog/audio-datasets#streaming-mode-the-silver-bullet).
Loading a dataset in streaming mode loads individual samples of the dataset at a time, rather than downloading the entire
dataset to disk:

```python
from datasets import load_dataset

dataset = load_dataset("distil-whisper/spgispeech", "L", streaming=True)
# take the first sample of the validation set
sample = next(iter(dataset["validation"]))
```

## Distil Whisper Usage

To use this dataset to reproduce a Distil Whisper training run, refer to the instructions on the
[Distil Whisper repository](https://github.com/huggingface/distil-whisper#training).

## License

This dataset is licensed under custom terms. To view the custom license for this dataset, refer to the original [dataset card](https://huggingface.co/datasets/kensho/spgispeech).

license: 其他许可
task_categories:
- 自动语音识别（automatic-speech-recognition）
language:
- 英语（en）
extra_gated_prompt: |-
您对标普全球（S&P Global, Inc.）子公司Kensho Technologies, LLC（以下简称“Kensho”）提供的Kensho转录数据集（以下简称“本内容”）的访问与使用，应受以下使用条款（以下简称“本条款”）约束。仅当用户接受并确认遵守本条款后，方可获得本内容的访问权限（此类用户统称为“授权用户”）。请在本条款末尾签署电子签名，以此声明您为授权用户，且已接受本条款并同意受其约束。
若您不愿受本条款约束，则不得使用本内容。请在使用本内容前仔细阅读本条款。
第1条 本内容
1.1 本内容仅可用于学术研究与内部使用，严禁用于以下行为：组装或创建数据库；开发或协助开发与本内容构成竞争关系的产品；识别、尝试识别或联系任何个人；或链接至其他数据集。
本内容包含公开业绩电话会议的音频及对应文本格式数据，以及所有配套衍生产品，其知识产权归Kensho及其第三方内容提供商所有。除本条款明确允许的情形外，您不得修改本内容、基于本内容创建衍生作品、重写或再处理本内容。未经Kensho事先书面同意，您不得向任何主体发布、展示、传输或重新分发本内容的全部或部分内容，或其衍生版本。您同意不就本内容中包含的个人信息联系Kensho或其关联方。
1.2 免责声明。您直接或间接通过本内容获取的所有内容，均未经过Kensho的审核或监控。Kensho不对任何此类内容的准确性、有效性、时效性、完整性或持续可用性作出任何明示或默示的保证、陈述或担保。
本内容仅为方便您使用而提供，并非对其中包含的观点或信息的重新发布或确认。Kensho或其第三方内容提供商无义务审核本内容，亦不对您使用本内容所产生的任何责任或义务承担责任。Kensho不对任何本内容的准确性、有效性、时效性、完整性或持续可用性作出任何明示或默示的保证或陈述，亦不对因依赖相关信息而产生的任何错误、延误或行为承担责任。此外，本内容仅代表发布当日的状态，且基于可能包含前瞻性预测或其他前瞻性陈述的电话会议内容。您不应依赖本内容来表达Kensho的观点或代表当前信息。Kensho及其第三方内容提供商均未承诺，也无义务更新任何本内容或就本内容的任何变更向您提供通知。
1.3 第三方内容的所有权。您确认，本内容中属于Kensho或第三方内容提供商的所有专有权利，仍归Kensho或该等第三方内容提供商所有，您仅享有按照本条款使用该等第三方内容的权利。本条款未授予的任何额外权利，需与Kensho另行签订直接协议方可获得。您确认，由Kensho或其第三方内容提供商编译、制备、选择和整理的本内容及第三方内容，是Kensho及其第三方内容提供商投入大量时间、精力和资金的成果，构成Kensho及该等第三方内容提供商的宝贵商业财产和/或商业秘密。Kensho保留美国及其各州版权法、商标法、服务商标法、专利法及其他法律赋予的所有权利和救济措施，包括但不限于任何旨在保护专有或保密信息的法律。您同意不会移除或修改本内容或第三方内容中可能出现的Kensho或第三方内容提供商的任何版权声明、披露信息、免责声明或其他通知、商号或商标，且任何经允许的本内容或第三方内容的复制和/或分发，均应包含本内容或第三方内容中出现的此类声明和/或商标。未经Kensho或该等第三方内容提供商事先书面同意，您不得使用Kensho或第三方内容提供商的名称或商标。除本条款授予的权利外，本条款无意转让任何所有权、权利、头衔或权益。任何额外权利需与Kensho另行签订协议方可获得。
1.4 发布的指南。除本条款外，使用本内容时您还应遵守并遵循任何可能不时发布和修订的发布通知、指南或规则。本内容中的任何内容均不应被视为向任何司法管辖区内的任何人购买证券的推荐或要约，或出售证券的要约。
1.5 注册数据。作为使用本内容的对价，您和/或您的雇主同意：(a) 在访问本内容前填写的注册表单提示下，向Kensho提供真实、准确、最新且完整的注册数据（定义见第3.1节）；(b) 维护并及时更新注册数据，确保其始终真实、准确、最新且完整。
1.6 终止用户访问的权利。Kensho保留随时自行决定限制、限制并立即终止您对本内容的全部或部分访问和使用的权利，且无需提前通知。
第2条 免责声明与责任限制
2.1 本内容按“现状”和“现有可用状态”提供，无任何形式的陈述或担保。使用本内容的风险由您自行承担。在任何情况下，Kensho或其第三方内容提供商均不对因依赖任何内容（包括第三方内容）而作出的任何决策、采取或未采取的任何行动承担责任，包括您对本内容的处理与存储。Kensho进一步明确放弃所有形式的担保，无论明示或默示，包括原创性、准确性、完整性、时效性、适销性、特定用途适用性及不侵权的担保。Kensho明确声明，且您放弃因您向第三方发布或提供本内容，或您向任何第三方作出的任何陈述或担保（无论是否与本内容相关）而产生的任何责任。Kensho、第三方内容供应商以及任何与Kensho合作的其他第三方，均不对因您使用本内容而导致或声称导致的任何损害或损失（包括直接损害、间接损害、附带损害、后果性损害及任何和所有其他形式的损害或损失，无论诉讼形式或索赔依据为何）承担直接或间接责任，即使Kensho或任何第三方内容供应商或其他与Kensho合作的第三方已被告知此类损害发生的可能性。
2.2 本内容并非旨在提供税务、法律、保险或投资建议，本内容中的任何内容均不应被解释为Kensho或任何第三方出售证券的要约、购买要约的招揽，或推荐任何证券。
2.3 对于第三方提出的任何索赔、诉讼、程序以及因任何性质的损失、损害、合理法律费用和其他合理开支产生的责任，您同意为Kensho及其关联方（包括其各自的董事、高管、员工和代理人）进行辩护、赔偿并使其免受损害，且该等责任全部或部分源于您对本内容的访问和/或使用、您未遵守本条款，或违反适用法律。
第3条 隐私
3.1 访问与收集。为访问本内容，在注册过程中，您或您的雇主需向Kensho提供特定信息，包括您的姓名、雇主或学术机构以及电子邮箱（以下简称“注册数据”）。此外，当您请求或查看本内容时，Kensho可能会获取与您对本内容的请求或访问相关的用户可识别信息（以下简称“访问数据”）。例如，当您访问本内容时，我们的Web服务器可能会识别您的：(a) 域名；(b) ISP的域名；(c) IP地址；(d) 浏览器类型；以及(e) 操作系统。如果您通过技术问题联系我们，我们可能会收集有关您系统的特定信息，包括：(a) 浏览器类型、版本和设置（例如Java和Cookie设置）；(b) 连接信息（例如SSL/HTTPS兼容性、带宽容量）；以及浏览器插件信息（例如您是否安装了Adobe、您的媒体播放器是什么、您能否打开Flash文件等）。
3.2 您的信息的使用。Kensho可将注册数据和访问数据用于研发目的、与用户沟通以及排查与本内容相关的任何技术问题。您确认，若需另行签订单独协议，Kensho可将注册数据与其关联方（定义见下文）共享。
3.3 您的信息的披露。除本条款另有说明外，Kensho不会在未经您许可的情况下披露、出租或出售从您处或关于您收集的个人信息。为上述条款中指定的目的，我们可将注册数据和访问数据转让或披露给标普全球公司及其关联方（以下简称“Kensho关联方”）以及受委托代表Kensho提供服务的第三方，例如协助Kensho向您提供本内容以及本内容中包含或可通过本内容访问的特定功能的第三方。我们也可出于向您提供本内容访问权限的目的，向Kensho关联方和第三方披露注册数据和访问数据。向此类第三方的披露将受保密协议约束，且在适用情况下受合同约束。Kensho也可能被要求向政府、监管或自律实体或机构披露信息，以响应监管询问或遵守适用法律、规则、条例、命令、传票或其他法律程序。
3.4 同意。通过(a) 同意本条款，或(b) 使用本内容，并在任何一种情况下提供我们上述规定的可能需要、请求或以其他方式收集的任何信息，您自由同意Kensho在美国和其他国家及地区为本条款规定的目的处理您的信息，且您也同意为上述目的将您的信息转移至任何第三方内容提供商（无论该实体目前位于何处）以及上述条款中描述的任何第三方，并符合适用的法律法规。如果您不允许Kensho收集您的任何信息或不同意本条款的任何条款，您不应使用本内容，并应退出本页面和/或本内容（视情况而定）。如果在向Kensho注册后，您希望撤回本第3.4条中授予的同意，即Kensho今后不再使用您的信息，您必须以书面形式通知Kensho，地址见第3.8节，并立即停止使用本内容。
3.5 咨询。如果您对本条款或我们持有的您的信息有任何疑问，请以书面形式联系Kensho。如果我们收到关于我们持有的您的个人信息的请求，我们将使用合理手段向您提供我们可以合理汇编的此类信息。您将有机会更正此类信息中的任何不准确之处。
3.6 加密。Kensho可能使用加密技术来保护与本内容相关的某些数据传输，但除非本内容另有说明，否则通过电子邮件或其他通信方式与本内容之间的通信均未加密。因此，您不应通过电子邮件向Kensho发送任何个人或识别信息，例如账户号码、信用卡号码、社会保障号码、密码等。通过使用电子邮件或其他电子通信方式，您确认您对通过此类方式发送的信息没有隐私期望，且Kensho不对第三方拦截此类发送的信息而导致的任何损失或损害承担责任。
3.7 联系信息。如果您对本使用条款、本隐私声明有任何疑问，或对我们持有的您的信息提出任何请求或查询，您可以通过以下方式以书面形式联系我们：发送邮件至[e-mail]，或邮寄至Kensho Technologies LLC，收件人：总法律顾问，地址：55 Water Street, New York, NY 10041。
第4条 其他条款
4.1 完整协议。本条款构成双方就本主题事项达成的完整协议，并取代双方此前就本主题事项达成的所有书面和口头协议及承诺。
4.2 可分割性。如果本条款的任何条款或其他规定被任何法律或公共政策认定为无效、非法或无法执行，则本条款的所有其他条款和规定仍应保持完全效力，只要本条款所设想的交易的经济或法律实质未对任何一方造成重大不利影响。
4.3 管辖法律；管辖法院。本条款应在所有方面受纽约州法律管辖，因本条款产生或与本条款相关的任何诉讼，均应在纽约州纽约县有管辖权的州或联邦法院提起。
4.4 放弃陪审团审判。在适用法律允许的最大范围内，您放弃就因本条款产生、根据本条款或与本条款相关的任何直接或间接诉讼或程序进行陪审团审判的任何权利。
4.5 冲突。如果本使用条款与任何与第三方内容相关的其他与Kensho签订的协议之间存在冲突，则以更严格的条款为准。

extra_gated_fields:
全名: 文本框
电子邮箱: 文本框
所属机构: 文本框
我接受本使用条款: 复选框
---

# 蒸馏Whisper：带时间戳的SPGISpeech数据集
本数据集是[SPGISpeech](https://huggingface.co/datasets/kensho/spgispeech)数据集的变体，新增了伪标注的Whisper转录结果，并与原始数据集元素一同返回。该伪标注转录结果由Whisper [large-v2](https://huggingface.co/openai/whisper-large-v2)模型采用贪心采样（greedy sampling）与时间戳预测方式，对输入音频数据进行标注生成。如需了解原始数据集的构建细节，请参阅原始[数据集卡片](https://huggingface.co/datasets/kensho/spgispeech)。

## 独立使用方法
首先，请安装最新版本的🤗数据集（Datasets）工具包：
bash
pip install --upgrade pip
pip install --upgrade datasets[audio]

可通过[`load_dataset`](https://huggingface.co/docs/datasets/v2.14.5/en/package_reference/loading_methods#datasets.load_dataset)函数将数据集下载并预处理至本地磁盘：
python
from datasets import load_dataset

dataset = load_dataset("distil-whisper/spgispeech", "L")
# 取验证集的第一个样本
sample = dataset["validation"][0]

您也可通过数据集工具包的[流式加载模式](https://huggingface.co/blog/audio-datasets#streaming-mode-the-silver-bullet)直接从Hugging Face Hub流式读取数据集。流式加载模式会单次加载单条数据样本，而非将整个数据集下载至本地：
python
from datasets import load_dataset

dataset = load_dataset("distil-whisper/spgispeech", "L", streaming=True)
# 取验证集的第一个样本
sample = next(iter(dataset["validation"]))

## 蒸馏Whisper训练使用方法
如需使用本数据集复现蒸馏Whisper的训练流程，请参阅[蒸馏Whisper代码仓库](https://github.com/huggingface/distil-whisper#training)中的说明文档。

## 许可证
本数据集采用自定义许可协议。如需查看本数据集的自定义许可条款，请参阅原始[数据集卡片](https://huggingface.co/datasets/kensho/spgispeech)。

原始信息汇总

Distil Whisper: SPGISpeech With Timestamps

数据集概述

这是一个基于SPGISpeech数据集的变体，增加了伪标记的Whisper转录本以及原始数据集元素。伪标记转录本是通过使用Whisper large-v2模型对输入音频数据进行贪婪采样和时间戳预测生成的。有关原始数据集的详细信息，请参阅原始数据集卡片。

独立使用

安装依赖

首先，安装最新版本的🤗 Datasets包：

bash pip install --upgrade pip pip install --upgrade datasets[audio]

加载数据集

数据集可以使用load_dataset函数下载并预处理到磁盘：

python from datasets import load_dataset

dataset = load_dataset("distil-whisper/spgispeech", "L")

获取验证集的第一个样本

sample = dataset["validation"][0]

流式加载

数据集也可以直接从Hub使用Datasets的流式模式进行流式加载。流式加载模式一次加载数据集的单个样本，而不是将整个数据集下载到磁盘：

python from datasets import load_dataset

dataset = load_dataset("distil-whisper/spgispeech", "L", streaming=True)

获取验证集的第一个样本

sample = next(iter(dataset["validation"]))

Distil Whisper 使用

要使用此数据集重现Distil Whisper训练运行，请参考Distil Whisper仓库中的说明。

许可证

此数据集受自定义条款许可。要查看此数据集的自定义许可证，请参阅原始数据集卡片。