Amazon Technologies, Inc.

Network Appliance, Inc.

Mohit Gupta

Jeetendra G Mirchandani

Leslie Johann Lamprecht

John David Dunagan

Malcolm Featonby

John Merrill Phillips

Brian James Schuster

Mihir Sadruddin Surani

Yuxuan Liu

Letian Feng

Nagender Parimi

Simon C Vining

Sravan Babu Bodapati

Katrin Kirchhoff

Roberto Pentz De Faria

Vivek Govindan

Liang Ge

Desai Nishitkumar Ashokkumar

Les Johann Lamprecht

Matias H Ganc

Rahul Gupta

Monica Lakshmi Sunkara

Ethan John Faust

Varun Sembium Varadarajan

Anureita Rao

Pu Paul Zhao

Asem Rustum

Srikanth Ronanki

Qijia Chen

Julian E S Salazar

Bo Pang

Jeffrey John Farris

Bin Xu

James Rigby

Jiaming (Jason) Sun

Wenbo Bu

Christian Egon Berkhoff Dossow

Himalay Mohanlal Joriwal

Sai Madhuri Bhavirisetty

Abhinav Kumar

Orestis Lykouropoulos

Akshay Nalwaya

Liangwei Guo

Yibin Wang

Bharath Raghavendran

K P N V D S Siva Rama

Hemant Kumar Gupta

Calvin Xuan Li

Naman Chhabra

Silviu Mihai Burz

Yide Zou

Swagath Kannan

Erik Quanstrom

Irfan M Mohiuddin

Techniques for personalized batch and streaming speech-to-text transcription of audio reduce the error rate of automatic speech recognition (ASR) systems in transcribing rare and out-of-vocabulary words. The techniques achieve personalization of connectionist temporal classification (CT) models by using adaptive boosting to perform biasing at the level of sub-words. In addition to boosting, the techniques encompass a phone alignment network to bias sub-word predictions towards rare long-tail words and out-of-vocabulary words. A technical benefit of the techniques is that the accuracy of speech-to-text transcription of rare and out-of-vocabulary words in a custom vocabulary by automatic speech recognition (ASR) system can be improved without having to train the ASR system on the custom vocabulary. Instead, the techniques allow the same ASR system trained on a base vocabulary to realize the accuracy improvements for different custom vocabularies spanning different domains.

Personalized batch and streaming speech-to-text transcription of audio

Features are disclosed for correlating a workload type with particular volume characteristics for a block storage volume. The volume characteristics may include a durability or a performance consistency for a particular volume. A computing device can obtain a set of workload parameters indicating a workload associated with a particular block storage volume. Based on the set of workload parameters, the computing device can determine a workload classification that links the set of workload parameters to a set of volume characteristics. The computing device can further compare the set of volume characteristics with the current set of volume characteristics for the block storage volume. Based on comparing the sets of volume characteristics, the computing device may determine a recommendation for a user. The computing device can dynamically modify the block storage volume based on the recommendation.

Dynamically modifying block-storage volumes using workload classifications

Features are disclosed for forecasting a usage of a block storage volume with a first configuration by a user. A computing device can forecast the usage of the block storage volume based on the historical usage of the block storage volume by the user. The computing device can determine additional potential configurations of the block storage volume. The computing device can further simulate the additional potential configurations of the block storage volume based on the forecasted usage of the block storage volume. The additional potential configurations may include a volume type, a volume size, or other volume characteristics. Based on the simulations of the additional potential configurations, the computing device may determine a recommended configuration. The computing device can dynamically modify the block storage volume based on the recommended configuration of the block storage volume.

Dynamically modifying block-storage volumes using forecasted metrics

A transcription service may receive a request from a developer to build a custom speech-to-text model for a specific domain of speech. The custom speech-to-text model for the specific domain may replace a general speech-to-text model or add to a set of one or more speech-to-text models available for transcribing speech. The transcription service may receive a training data and instructions representing tasks. The transcription service may determine respective schedules for executing the instructions based at least in part on dependencies between the tasks. The transcription service may execute the instructions according to the respective schedules to train a speech-to-text model for a specific domain using the training data set. The transcription service may deploy the trained speech-to-text model as part of a network-accessible service for an end user to convert audio in the specific domain into texts.

Model training system for custom speech-to-text models

A database service that receives and manages time series data into a set of generational data tables. The database service generates a set of generational data tables configured to receive a subset of time series data. Each generational data table is created with an initial set of access parameters that facilitates the receipt of incoming time series data. Thereafter, the database service creates additional generational data tables to receive the continuous received time series data and then modifies the access parameters of the previously created data table, such as for read only rights.

Generational databases for managing time series data

Techniques for a service provider network to allow users to quickly and easily establish a testing environment to test various virtual machine (VM) instance types for hosting their workloads. Rather than identifying and recommending optimized VM instance types for hosting workloads of users, the techniques allow for users to initially test the VM instance types and determine how well their workloads perform on the VM instance types. Users can quickly and easily (e.g., 'one-click' input) request that a testing environment be established. The optimization service can then test one or more recommended VM instance types for the users' workloads in the testing environment. The optimization service can monitor the performance of the VM instance types while they host the 'test workloads,' and provide the users with performance metrics to help them decide if they would like to migrate their workloads to the recommended VM instance types.

Testing recommended compute platforms for a workload

A virtual private cloud (VPC) that includes one or more computing devices (e.g., a physical computing device, a virtual computing device, etc.) that each implement a service present in an actual production environment is provided herein. For example, at the request of a user, an instance deployment manager may replicate one or more of the services provided by an actual production environment such that the services can be executed by the computing devices within the VPC. The computing devices within the VPC may be configured to communicate with each other. However, the computing devices may not communicate with devices outside the VPC. Thus, the VPC may represent a sandboxed or isolated test stack that allows a user to independently test code within a replicated production environment.

Automatic test stack creation via production system replication

Techniques for managing dynamically scalable virtualized compute instances within a provider network are described. A dynamically scalable instance has a baseline performance level and a maximum performance level associated with a computing resource, such as a processor, a memory, a network interface, etc. In response to receive a request to launch a dynamically scalable instance, one or more services select a computer system to host the requested instance from a pool of instance-hosting computer systems of a provider network. The instance-hosting computer systems include a monitoring agent that reports computing resource usage data to the one or more services. The one or more services monitor the resource usage of the instance-hosting computer systems and migrate instances so that the dynamically scalable instances can operate up to the maximum level of performance.

Adjusting dynamically scalable instance hosting based on compute resource usage

Techniques for managing compute resource usage by virtualized compute instances within a provider network are described. A computer system connected to a provider network hosts one or more compute instances. An agent associated with the computer system obtains per-compute instance resource usage information of computer system resources such as processors, memory, and network interfaces. The agent sends the usage information to a usage management service of the provider network. The usage management service generates usage limits based on the usage information from the agent and on usage targets and sends the usage limits to the computer system. The computer system limits the resource usage of the compute instance based on the usage limits.

Managing compute resource usage based on prior usage

Architectures and techniques to store and search logging information are provided. In some embodiments, statements included in the logging information can be categorized according to respective unique identifiers. A category of statements can be retained in one or more files within a directory associated with a unique identifiers representative of the category. The directory can be included in a distributed storage system. In addition, metadata including the unique identifier and a defined searchable field or another defined discrete amount of information pertaining to a statement associated with the unique identifier can provide an indexation of an available category of statements. In addition or in other embodiments, an interface unit can permit querying the logging information via programmatic queries or other queries received via a user interface. The query can include a desired search pattern and a response to the query can include a list of one or more unique identifiers indicative of respective statements matching the search pattern. Logging information statements sorted according a defined criterion can be provided in response to the selection.

Architecture and techniques to search logging information

The present disclosure involves ensuring that items sold by a merchant appear in certain network pages. A network page is automatically accessed in a computing device. The network page is generated by a referral network site that displays at least one offering of an item by at least one competitor merchant. A unique identifier associated with the item is identified from the network page in the computing device. The unique identifier is stored in association with the item in a data store accessible to the computing device. The item and the unique identifier are included in a feed sent from the computing device of a merchant to the referral network site. The feed includes a listing of the item and a plurality of other items, wherein the item and the other items are offered for sale by the merchant.

Ensuring appearance of merchant offering in network sites

This document relates to classification of items included in a data feed. For example, a feed of items is generated for provision to a referral network site. The feed includes some of the items sold through an electronic commerce network site. Item classification information corresponding to an item sold through the electronic commerce network site is identified based upon the taxonomy information of the referral network site and included in the generated feed of items.

System, method, and computer readable medium for item feed classification

Disclosed are various embodiments for generating product feeds associated with a product catalog for a third party site. Entries in a product feed that are associated with products that are in turn associated with a price that is below a threshold specified by a minimum advertised pricing restriction can be flagged. A hyperlink can be provided through which a user can add the product to a shopping cart in an electronic commerce system, which can allow the third party site to display pricing information for the product.

Generating product feeds

Growing community of inventors

Bellevue, WA, United States of America

Mohit Gupta