Hi. I'm Patty Raymond for Amazon Web Services, Training and Certification. For the next few minutes, I'm going to walk you through a quick overview of Amazon Transcribe. Amazon Transcribe is an automatic speech recognition service also known as ASR, that's designed to make it easy for developers to incorporate speech-to-text capabilities into their applications. In this course, I'll highlight the features of Amazon Transcribe, talk a bit about how it works, and provide a few examples of common use cases. I'll also tell you how to get started and give a quick demonstration. Historically, people saved most data in structured formats that were relatively easy to search. For example, information entered into databases via forms or saved and spreadsheets is pretty easy to analyze even at very high volumes. Increasingly, we capture data in less structured formats that are more difficult to analyze on a large scale. In particular, the speech in audio and video files needs to be recorded as text before it can be searched or categorized. Traditionally, converting audio content to text has required a lot of human intervention. Someone has to play back the recording and capture all of the speaker's words in text form and then review and edit the transcribed content. Even with the benefit of a speech-to-text tool, a lot of editing and formatting is required. This is particularly true for outputs that require high levels of accuracy and readability like video subtitles. The high level of effort to manually convert these media to text means that only the most critical items are transcribed and valuable information may be overlooked. Amazon Transcribe can help you address these potential barriers. Amazon Transcribe is a fully managed and continually trained ASR service that accepts common audio formats including WAV, MP3, MP4, and FLAC, and can accurately transcribe both low fidelity and high fidelity audio. This makes it a natural fit for use cases such as transcribing customer service calls into analyzable data or generating subtitles for videos with a high level of accuracy. Amazon Transcribe uses deep learning to provide accurate and quick transcriptions. It was designed to make it easy for you to get even greater accuracy and usefulness from the transcribed output. For example, the output is not just one long uninterrupted string of text. Instead, Amazon Transcribe uses machine learning to add in punctuation and grammatical formatting. So the texts you get back is immediately usable. Amazon Transcribe also timestamps every word, which makes it easy to align subtitles for close captioning. It also makes it easy to index. Transcripts include confidence levels for each word of the transcribed output, making it easy to determine where more editing or quality assurance may be required. Amazon Transcribe also helps you mitigate issues with source audio to improve accuracy. For example, you can upload custom vocabulary that can help improve the accuracy if the content you're transcribing has industry-specific terms like medical or legal terms. You can choose the language of the source file. The supported languages are listed, and Amazon Transcribe attribute speech to different speakers to make it easier to interpret the output. Because Amazon Transcribe was designed to be easy for developers to use, there are only a few steps to get started. You don't need to understand how the underlying models work and you don't need to build your own machine learning models. You invoke the service via API and with a few lines of code, you're ready to transcribe files on Amazon S3. You can easily initiate the transcription operation via the AWS management console and you can test the service via the console too. Amazon Transcribe offers secure communication between the service and your applications through SSL encryption. You can access your transcriptions via assigned URLs. You can also make sure that information is kept secure and confidential by controlling access to Amazon Transcribe through AWS IAM policies. The service integrates easily with other AWS services such as Amazon Comprehend, Amazon ES, Amazon Translate, and Amazon Athena. You can easily leverage powerful deep learning models combined with highly scalable index searching and analysis tools to reap high-value from your audio and video content. Here are a few examples. A call center uses Amazon Transcribe to produce transcriptions of low-fidelity phone calls with high accuracy. The center uses Amazon Comprehend on the transcribed data to identify key phrases and customer sentiments so that they can classify the calls. They use Amazon Athena to query the data and Amazon QuickSight to visualize the results and analyze trends. The Human Resources Department uses Amazon Transcribe to generate transcriptions of meetings and training sessions, then indexes the results using Amazon ES. Internal stakeholders can quickly search for content on demand to locate topics of interest. A video production business uses Amazon Transcribe to produce transcriptions of audio tracks for subtitles. Limited editing is done on the output files and the quality assurance team can focus on sections that have lower confidence ratings. Using the timestamps in the transcript file, they can easily align the captions with the video. They use Amazon Translate to convert the transcript into additional languages. You can even translate conversations. Use Amazon Transcribe to generate the transcription of the speaker's message. Use Amazon Comprehend to recognize the language of the speaker, and use Amazon Translate to translate the message into the receiver's language. Then, use Amazon Polly to read the translated message to the recipient. To get started, you'll need to save the file you want to transcribe on Amazon S3 stored in a bucket with the proper permissions. The files must be in one of four formats; FLAC, MP3, MP4, or WAV, and they must be less than two hours long. For best results, source files should use lossless compression, FLAC or WAV, and recording should be done in a low-noise environment. You'll want to limit crosstalk too. You'll need an AWS account with an IAM user who has full access to the transcribed API calls, and you should be familiar with the CLI. You'll also need a text editor. We also recommend that you review the developer's guide which you can find in the AWS documentation under Amazon Transcribe. For the demonstration, I'll use the AWS management console to start a job, verify that it's completed, and then review the results. On the AWS management console, select "Amazon Transcribe." Then click the "Try Amazon Transcribe" button to open the "Create Transcription Job" dialogue. Give your job a relevant name. For this demonstration, I'm going to use Gettysburg Address test. Paste or type the location of the source file on Amazon S3 that you want to transcribe. Select the source language. For this demonstration, I'll use English, and select the format of your source file. For this example, it's MP3. Optionally, I can specify the media sampling rate. This rate must match the sampling rate of the associated media content. Sampling rates between 8000 hertz and 48,000 hertz are supported by Amazon Transcribe. Once I fill that in, I click the "Create" button and my job has been submitted for processing. Once the request has been submitted, I can check the status of the job by selecting "Transcription Jobs" from the menu on the left. This opens a list of available jobs. You may see older transcripts. That's because the service makes transcription results available for 90 days. In the status column, I can see that my job is in progress. If I click on the job name itself, I can review the details of the job and also review or copy the sample JSON request for this job. When the status of the job is completed, I'll be able to select the output URL from the job details page and review the transcribed text. The output file is placed in an S3 bucket and is only accessible with assigned URI. For added security, the URI is valid for only a few minutes after you initially request the job results. Just taking a quick look at the transcript, I can see that the content of my MP3 file has been transcribed with formatting and punctuation, and I can see the timestamps and confidence levels for each word in the file. In summary, Amazon Transcribe is a fully managed and continually trained ASR service that is designed to make it easy for developers to incorporate accurate speech-to-text capabilities into their applications. I hope you found this introduction to Amazon Transcribe helpful, and that you're ready to try it yourself. Please continue to explore other related courses. I'm Patty Raymond for Amazon Web Services, Training and Certification.
