AI Modules

DataScouting creates AI modules to solve real world problems.
Our technologies help companies and organizations manage their structured
and unstructured data and extract meaningful information from them.

AutoSpeech transforms speech into text including additional metadata, such as per word time synchronization and speaker detection. It can be installed on-premises working fully offline or can be used on the cloud as SaaS. AutoSpeech is built using state-of-the-art deep learning neural networks, supporting transfer learning across languages, making the creation of new language models easier than ever. Currently, AutoSpeech supports 42 languages, while additional languages can be supported, both including specialized vocabularies, such as medical or judicial terms, or focusing on particular domains, such as broadcasting or call centers. AutoSpeech guarantees maximum accuracy of transcriptions as well as speed of speech recognition outputs.

Autospeech Editor includes a user interface for manually editing transcriptions keeping the words synced in time. When using the Editor, you can very easily create subtitles, conference proceedings, or minutes of meetings. When using speech recognition and editing transcriptions of audio files you can minimize the work/effort involved in transcription workflows by more than 50%.

DataScouting’s voice biometrics system uses unique voice characteristics for identity verification. It includes two main modules: gender identification and speaker classification. The gender identification module uses advanced machine learning to determine if a speaker is male or female by analyzing acoustic features. The speaker classification module identifies speakers in conversations or recordings by comparing vocal characteristics with a database of known speakers. Together, these modules provide accurate speaker identification and authentication. Applications include transcription, translation, voice assistants, virtual reality, forensic analysis, and security systems.

FaceScouting is an innovative software solution for face recognition in video streams using Deep Learning technologies. FaceScouting guarantees maximum accuracy and works using CPUs or GPUs to guarantee efficiency. By providing photos of persons of interest, FaceScouting will create a new model, identify the persons in the streams, and push the results back to your application. FaceScouting integration is accomplished via a RESTful API. It can be installed on-premise, working fully offline or can be used on the cloud as SaaS.

LogoScouting is an innovative software solution for logo recognition in video streams using Deep Learning technologies. LogoScouting guarantees maximum accuracy and works using CPUs or GPUs to guarantee efficiency. By providing only a few photos of the logos of interest, LogoScouting will create a new model, identify the logos in the streams, and push the results back to your application. The system is built to identify logos in sports games, print on items or even on sports outfits. LogoScouting integration is accomplished via a RESTful API. It can be installed on-premise, working fully offline or can be used on the cloud as SaaS.

AdScouting automatically identifies advertisements or any repeated clip in live audio or video streams, guaranteeing success rates up to 99.9%. Based on neural networks and proprietary technology, AdScouting can be integrated in any application via a RESTful API or SDK and can be used either as SaaS or installed on-premises. Upload audio files, provide stream URLs, let the system push ad detections to your application or query its database.

Summarizer uses deep neural networks to understand the most important parts of a document and create a short summary. The summary is not just a selection of sentences from the original text. Summarizer does not create a simple abstract but provides the main idea, which will support the overall purpose of the entire article. Originally trained with terabytes of media data, Summarizer offers maximum performance and works with cloud installations or on-premises.

Automated Sentiment Analyzer is a bespoke solution that can be trained with your data according to your definition of positive, neutral or negative. Sentiment analysis is a difficult task to tackle, especially since rating the polarity of a text (positive, negative, neutral) can prove particularly difficult to conduct even for human analysts. DataScouting offers state-of-the-art generic sentiment analyzers, and specializes in creating customer centric sentiment analysis by creating models that identify pro vs anti government content, buy vs sell sentiment signals, etc.

Financial Sentiment Analyzer is a state-of-the-art sentiment analyzer that specializes on identifying buy vs sell sentiment in textual content on social media. Trained with thousands of human-annotated data, Financial Sentiment Analyzer offers unprecedented accuracy, while DataScouting’s research team keeps updating the models on a bimonthly basis.

We work with open-source Optical Character Recognition systems, and we have created custom models for layout and text recognition for several of our clients across the globe for different domains and languages. Supporting more than 200 different language models and being able to train the models using customer data, DataScouting provides OCR solutions that offer the maximum accuracy for different use cases, such as newspapers, magazines or older books that may include scanned inaccuracies or present other problems.

News Ticker Extractor processes video files, identifies any text appearing, and extracts this text with time stamps (optional). The system can also process news tickers or rolling subtitles, and it works with both left to right and right to left languages.

Our Deep Neural Network models automatically identify hate speech in social media networks in five languages. Hate speech and online toxicity is an important topic, and identifying it as soon as possible is of utmost importance. DataScouting’s Hate Speech Recognizer has been trained with hundreds of thousands of social media data, manually annotated, to provide the greatest accuracy. It provides users with: toxicity scores to enable them to mute and block abusers at scale and the ability to create reports with any harmful comments and accounts.

Entity Recognizer a) automatically identifies organization names, place names, and surnames from textual content, b) enriches the text metadata, and c) helps to identify hidden connections between documents. The system performs entity recognition by using neural networks that identify the syntax behind each language without providing the system with a list of place names or surnames.