Prepare and test your skills

Prepare and test your skills

Evaluate Audio Input and Output Scenarios means choosing the correct pre-trained speech API based on the direction of conversion. Speech-to-Text (opens in a new tab) converts spoken audio into written text for input scenarios, while Text-to-Speech (opens in a new tab) converts written text into natural-sounding speech for output scenarios. Both APIs are pre-trained, so developers can add speech recognition or voice generation to applications without training a custom model.
Speech-to-Text is the pre-trained Google Cloud API that integrates speech recognition into developer applications. Audio flows from a spoken source, such as a voicemail recording or customer call, to the Speech-to-Text API, and written text flows back from the API as the result. A common business use is transcribing a voicemail recording into a text message that can be searched, stored, or analyzed for records and follow-up.
Speech-to-Text can also run at scale from BigQuery (opens in a new tab) through BigQuery ML. Data flows from audio files stored in object tables through the ML.TRANSCRIBE function to text results stored in BigQuery, with all inference occurring in Vertex AI Agent Platform (opens in a new tab). The workflow depends on a remote model that represents the API endpoint, so first create the remote model with CLOUD_AI_SPEECH_TO_TEXT_V2 as the service type and optionally specify a recognizer to process the audio, then run ML.TRANSCRIBE against that model. Choose Speech-to-Text when the business need starts with voice and ends with text, including transcription, captions, and voice-driven workflows.
Text-to-Speech is the pre-trained Google Cloud API that converts written text into natural-sounding speech. Text flows from an application or document into the Text-to-Speech API, and audio flows out from the API as spoken output. The service depends on the input text, so the clarity of the voice response follows directly from the clarity of the source text. Choose Text-to-Speech when the business need starts with text and ends with voice, including automated voice responses such as phone assistants and notifications, accessibility support that reads content aloud, and real-time voice interaction where users need to hear information.
Choosing Between Audio APIs means matching the conversion direction to the starting input and required output. The two audio APIs are opposites in direction and purpose, so the starting input determines the correct choice:
| API | Conversion direction | Choose when |
|---|---|---|
| Speech-to-Text | Audio to text | The input is spoken audio and the required output is written text, such as voicemail to message |
| Text-to-Speech | Text to audio | The input is written text and the required output is spoken audio, such as an automated voice reply |
When a solution needs both directions, the APIs compose into a voice loop where Speech-to-Text converts the customer's spoken request to text for processing, then Text-to-Speech converts the text answer back to speech for the customer to hear.
Exam tip: Speech-to-Text converts audio to text, while Text-to-Speech converts text to audio — check the starting input to pick the correct API.
Assess Visual Content Requirements means selecting the correct Cloud Vision API (opens in a new tab) feature for an image task. The Cloud Vision API is a pre-trained Google Cloud API that analyzes images without requiring custom model training. An application sends an image to the API with a requested feature type, such as label detection or text detection, and the API returns annotations that describe what it found. Images can come from a local file sent as a base64-encoded string or from a file stored in Cloud Storage (opens in a new tab), which makes the API a fit for photos, scans, product images, and user-uploaded pictures.
Image Labeling for Organization and Search is automatic description of image content to support organization and visual search. The LABEL_DETECTION feature examines the whole image and returns labels such as objects, scenes, or activities, which supports organizing photo libraries, tagging product catalogs, and improving visual search. Because the model is pre-trained, teams do not need to collect training data or build a classifier to get started. For custom image categories that pre-trained labels do not cover, AutoML Vision classification can be trained for classifying image content, but document classification should use text-based methods after OCR (opens in a new tab). Choose AutoML Vision classification when the decision depends on what the image looks like, and choose text-based classification with the Natural Language API or AutoML Natural Language when the decision depends on what the document says.
Reading Text in Images with OCR means converting visible text in images into machine-readable text. OCR in the Cloud Vision API supports operational needs such as extracting prices from shelf photos, reading serial numbers, digitizing receipts, and processing scanned forms. The API offers two main OCR feature types that split by how much text the image contains and how the image is stored:
| Feature type | Best for | Important behavior |
|---|---|---|
TEXT_DETECTION | Sparse text within a larger image, such as a sign, label, or street scene | Optimized for small areas of text inside a photo |
DOCUMENT_TEXT_DETECTION | Dense text, documents in PDF or TIFF format, and handwriting | Takes precedence when both TEXT_DETECTION and DOCUMENT_TEXT_DETECTION are present in the same request |
Exam tip: Choose TEXT_DETECTION for sparse text in a larger image and DOCUMENT_TEXT_DETECTION for dense documents, PDFs, TIFFs, or handwriting.
Detecting Objects, Faces, and Visual Details means finding and describing distinct visual elements and where they appear. The OBJECT_LOCALIZATION feature detects and extracts multiple objects in one image, which supports counting products on a shelf, checking inventory, or guiding warehouse work. The result identifies both what each object is and where it sits in the image, so downstream systems can act on each detected item. Related visual detection features in the Cloud Vision API cover other common requirements:
FACE_DETECTION detects faces within the image for photo organization or user interaction.LOGO_DETECTION detects company logos within the image for brand monitoring and product identification.LANDMARK_DETECTION detects geographic landmarks within the image for travel, real estate, or location tagging.WEB_DETECTION finds topical entities such as news, events, or celebrities and finds similar images on the web using Google Image Search.Additional image analysis features describe presentation rather than identity. IMAGE_PROPERTIES computes properties such as dominant colors, while CROP_HINTS suggests vertices for a crop region to improve thumbnails and displays.
Moderating Images for Safety and Compliance means screening images for potentially unsafe content before display, storage, or publication. The SAFE_SEARCH_DETECTION feature runs SafeSearch analysis on an image and flags content that may violate policy or be unsuitable for some audiences. The image flows from the application to the Cloud Vision API, and the moderation result flows back to the application, which then decides to allow, block, blur, or send the image for human review. Choose SAFE_SEARCH_DETECTION when the business need is compliance, brand protection, or safer customer-facing galleries, marketplaces, and user-uploaded content.
Analyze Text and Language Processing Use Cases means choosing the Google Cloud pre-trained API whose function matches the text job. Text processing sends text to an API and receives an analysis or translation back, and Google Cloud exposes these capabilities as separate services with their own endpoints. The Cloud Natural Language API (opens in a new tab) (language.googleapis.com) analyzes text, while the Cloud Translation API (opens in a new tab) (translate.googleapis.com) translates it. For translation work, the Cloud Translation API data-handling rules determine which use cases it can safely serve.
How Cloud Translation API Handles Text describes Google's narrowly limited handling of text sent for translation. When text is sent to the Cloud Translation API, it is held briefly in memory only long enough to perform the translation and return the results, rather than being stored on Google servers. The following rules define what Google does and does not do with that text:
Because the text is not retained or reused, a business can route customer-facing or sensitive text through the Cloud Translation API without the content becoming training data or public material. The customer keeps ownership of what it sends, and the only processing that occurs is the translation itself.
Translation Endpoints and Data Location determine whether translation data can be kept inside a particular region. Cloud Translation - Advanced API supports regionalized endpoints that can be configured for data location, while global endpoints cannot be configured to remain in a specific region for processing:
| Option | Regionalized endpoints | Global endpoints |
|---|---|---|
| Cloud Translation - Basic API | Not offered | Yes — data cannot be configured to stay in a specific region |
| Cloud Translation - Advanced API | Yes — configurable for data location | Yes — data cannot be configured to stay in a specific region |
| AutoML Translation | Not offered | Yes — data cannot be configured to stay in a specific region |
A business with a requirement to keep translation data inside a particular region therefore needs Cloud Translation - Advanced API with a regionalized endpoint. Choosing a global endpoint on any of the three options removes that control.
Exam tip: Cloud Translation - Advanced API is the only option that supports regionalized endpoints for data location; the Basic API and AutoML Translation offer only global endpoints.
Glossaries and Access Control govern how customer-specific translation terminology is stored and reached. Glossaries let a translation use customer-specific terminology, and the Cloud Translation API stores them encrypted and securely on Google Cloud. Glossaries comply with the Cloud Data Processing Addendum and are managed from the customer's own Google Cloud project. Access to glossaries is governed by fine-grained Identity and Access Management (IAM) permissions, so only principals holding the proper permissions can reach a glossary. This means the boundary around a glossary is set by the customer's own project and IAM policy rather than by the translation service alone.
What Cloud Translation API Does Not Permit is reselling translation itself as a service. A business may still integrate the Cloud Translation API into applications of independent value, so the restriction targets reselling translation rather than building a product. The distinction matters for product design because translation can be one feature inside a larger application, but the API cannot be repackaged and sold as a standalone translation service.
CLOUD_AI_SPEECH_TO_TEXT_V2 and running ML.TRANSCRIBE.TEXT_DETECTION is for sparse text in images, while DOCUMENT_TEXT_DETECTION is for dense documents, PDF, TIFF, and handwriting.SAFE_SEARCH_DETECTION screens images for unsafe content, and OBJECT_LOCALIZATION detects multiple objects and their locations.The Speech-to-Text API converts spoken audio into written text for input scenarios such as transcribing a voicemail recording into a searchable message, while the Text-to-Speech API converts written text into natural-sounding speech for output scenarios such as automated voice responses and accessibility support. Choose Speech-to-Text when the business need starts with voice and ends with text and choose Text-to-Speech when it starts with text and ends with voice.
TEXT_DETECTION is optimized for sparse text within a larger image such as a sign, label, or street scene, while DOCUMENT_TEXT_DETECTION is for dense text, documents in PDF or TIFF format, and handwriting. DOCUMENT_TEXT_DETECTION takes precedence when both feature types are present in the same request.
Cloud Translation - Advanced API is the only option that supports regionalized endpoints that can be configured for data location. Cloud Translation - Basic API and AutoML Translation offer only global endpoints that cannot be configured to stay in a specific region.
Text sent to the Cloud Translation API is held briefly in memory only long enough to perform the translation and return the results rather than being stored on Google servers. Google does not use the content for any purpose except providing the service, does not make it public or share it with third parties, does not use it to train or improve translation features, and does not claim ownership of it.
An e-commerce company wants to automate how it processes thousands of images uploaded daily by sellers and customers. The company needs a pre-trained Google Cloud service that can address the following business requirements without training custom machine learning models:
Which Google Cloud service should the company use?
Text-to-Speech API
Cloud Data Loss Prevention API
Cloud Vision API
Cloud Translation API
Last updated